<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:media="http://search.yahoo.com/mrss/"><channel><title>Application Development</title><link>https://cloud.google.com/blog/products/application-development/</link><description>Application Development</description><atom:link href="https://cloudblog.withgoogle.com/blog/products/application-development/rss/" rel="self"></atom:link><language>en</language><lastBuildDate>Wed, 18 Nov 2026 17:00:40 +0000</lastBuildDate><image><url>https://cloud.google.com/blog/products/application-development/static/blog/images/google.a51985becaa6.png</url><title>Application Development</title><link>https://cloud.google.com/blog/products/application-development/</link></image><item><title>Sashiko: Lessons from building an agentic code review system for Linux</title><link>https://cloud.google.com/blog/products/application-development/lessons-from-building-sashiko-an-agentic-code-review-system/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The speed of development in large open-source projects has always been limited by code review, not code generation. Even as coding agents get better at writing code, they do not accelerate development on their own. This is especially true of patches to the Linux kernel. Without a way to build trust in proposed changes, more generated code simply piles up in front of the same small group of human maintainers. Some of those patches are solid changes, just written faster; many others are plausible-looking "AI slop" sent by contributors who do not fully understand the systems they are modifying. And the influx of AI-generated security reports, many of which are duplicates or inaccurate, only adds to maintainers’ load.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Linux runs billions of Android phones, most cloud providers, and nearly every supercomputer. It spans more than 35 million lines of code, and over 85,000 changes land every year. Automated build bots and fuzzers like &lt;/span&gt;&lt;a href="https://github.com/google/syzkaller" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Syzkaller&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; catch build errors and crashes, but code review has always been the main way to guarantee the stability and safety of the kernel. Every day, maintainers read hundreds and hundreds of proposed patches by hand to catch memory leaks, deadlocks, race conditions, and security bugs.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-paragraph_with_image"&gt;&lt;div class="article-module h-c-page"&gt;
  &lt;div class="h-c-grid uni-paragraph-wrap"&gt;
    &lt;div class="uni-paragraph
      h-c-grid__col h-c-grid__col--8 h-c-grid__col-m--6 h-c-grid__col-l--6
      h-c-grid__col--offset-2 h-c-grid__col-m--offset-3 h-c-grid__col-l--offset-3"&gt;

      






  

    &lt;figure class="article-image--wrap-small
      
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_sIAzTvL.max-1000x1000.png"
        
          alt="1"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  





      &lt;p data-block-key="uxivt"&gt;To help maintainers handle the load and catch bugs before they land, the &lt;a href="http://aiinfra.google/"&gt;Google AI &amp;amp; Infrastructure&lt;/a&gt; team built &lt;a href="https://github.com/sashiko-dev/sashiko"&gt;&lt;b&gt;Sashiko&lt;/b&gt;&lt;/a&gt;, an open-source, agentic code review system powered by Gemini. We named the project after the Japanese word 刺し子, a stitching technique for artistically reinforcing worn cloth, and donated it to the Linux Foundation earlier this year.&lt;/p&gt;&lt;p data-block-key="854tn"&gt;In this blog, we cover the problem Sashiko was designed to solve, how it works, what the data shows, and what we learned from building and running it in the wild.&lt;/p&gt;&lt;h3 data-block-key="cng2g"&gt;&lt;b&gt;Reviewing code is hard&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="3trf0"&gt;Reviewing a change in a complex code base is objectively hard, both for humans and LLMs. It often requires a deep understanding of how the system works in general, well outside the proposed change. In fact, it’s pretty common to see patches that engineers claim perfectly solve the problem, but if they were to be merged, would actually introduce even worse problems (albeit subtle). Traditionally, only experienced maintainers have been able to spot and prevent these issues. The same is true for agentically generated code.&lt;/p&gt;&lt;p data-block-key="c5fuc"&gt;To verify the correctness of a change, a potentially large number of invariants should be proven, which is often a much larger and harder task than developing the change itself. This is true for both human and agentic reviewers. However, LLM-based reviewers have some unique properties, both positive and negative:&lt;/p&gt;&lt;ol&gt;&lt;li data-block-key="6pc49"&gt;&lt;b&gt;LLMs often struggle to put in enough effort.&lt;/b&gt; LLMs are trained to achieve the goal of spending fewer tokens. So they often tend to cut corners and dismiss concerns without proper investigation or flag false positives without proper due diligence.&lt;/li&gt;&lt;li data-block-key="7uvtf"&gt;&lt;b&gt;LLMs are easier to bias&lt;/b&gt;. The current generation of LLMs usually performs much better if provided a single relatively simple task (e.g., prove the patch for memory leaks) rather than multiple tasks (e.g., explore several complex, varied concerns). It’s pretty common for an LLM to find an obvious issue in the code and refuse to look deeper for other types of issues. Also, commit log text can sometimes affect the result of code review in a non-predicatable or even malicious way.&lt;/li&gt;&lt;li data-block-key="1a35f"&gt;&lt;b&gt;LLMs don’t have persistent memory.&lt;/b&gt; Surprisingly, this is a positive thing for code review: An LLM can review the same or similar patch multiple times as if it were a new patch. No sane human can review the same patch applied to 100 slightly different branches to find the few that have incompatible changes.&lt;/li&gt;&lt;/ol&gt;&lt;h3 data-block-key="bgi0a"&gt;&lt;b&gt;How Sashiko reviews a patch&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="6tlin"&gt;To avoid these failure modes, Sashiko separates &lt;b&gt;finding candidate bugs&lt;/b&gt; from &lt;b&gt;verification of those bugs&lt;/b&gt;, and requires each pass to record what it ruled out:&lt;/p&gt;
    &lt;/div&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;Gather context and plan\r\n  -&amp;gt; Parallel specialist sessions (concerns + dismissed concerns)\r\n  -&amp;gt; Triage\r\n  -&amp;gt; Per-bug deep checks\r\n  -&amp;gt; Review report&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd5ae6d90&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Notes on each step:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Gather context and plan:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Sashiko first pulls in the code around the diff (modified functions, struct definitions, and direct callers) and loads the subsystem's domain guides, so the changed lines are never judged in isolation. A planner then decides which specialist passes are needed (skipping hardware checks on pure software code, for example) and sets a reasoning budget for each pass based on patch complexity.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Parallel specialist sessions:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Up to seven specialist sessions run in parallel, each focused on a single class of problems: commit intent, implementation match, control flow, resource lifetimes, locking, security, and hardware. Giving each pass a single focus prevents one type of bug from crowding out another. Each session searches and reads the repository with git tools (averaging about 100 tool calls per review) and outputs two lists: concerns it believes are real, and candidate concerns it investigated and ruled out, along with the code that supports that conclusion.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Triage:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; A lead verifier audits those two lists together. Clear-cut bugs already proven by the code in front of it pass straight through; dismissals proven by a local check are dropped. Everything else deserves additional investigation. Crucially, that includes not only contested bugs where specialists disagreed, but also &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;shaky self-dismissals &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;— cases where a specialist spotted a real bug and dismissed it by assuming, without proof, that some unseen caller or lock would prevent it.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Per-bug deep checks:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Instead of checking all difficult cases in one shared prompt, Sashiko gives each difficult case its own verification agent that runs in parallel. Each agent gets one specific question to answer with git tools under a symmetrical rule: Neither a bug claim nor a dismissal is accepted without concrete code proof across &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;all&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; call sites. If a specialist talks itself out of a real bug based on a false assumption, this step catches it and promotes it back to a real finding.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Review report: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Sashiko generates a report that sounds like a human maintainer asking questions about the proposed change, matching the style expected on a Linux mailing list.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;All this happens at no cost for developers. Deep agentic reviews at scale require a substantial token budget. If developers had to pay for them with their own API keys, many important contributors would be pushed away. Instead, Sashiko reviews patches sent to public Linux kernel mailing lists, and Google runs the service at &lt;/span&gt;&lt;a href="https://sashiko.dev" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;sashiko.dev&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. Everything runs in the open: every prompt, tool call, reasoning step, and ruled-out bug is public. The project initially used the open-source Linux review prompts started by Chris Mason, and more than 100 contributors, including prominent kernel maintainers and engineers, have since helped to improve prompts and Sashiko itself. Recently, Sashiko switched to a different prompt system that combines per-tree or per-version auto-generated notes on code with subjective maintainers notes, which contain directions and instructions from maintainers that can’t be derived from the code.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;What the data shows&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The real test of an automated reviewer is whether maintainers pay attention to the reports it sends out, and whether those reports track actual software quality. Seven months of Sashiko production data from the kernel mailing lists give a clear picture of both these questions.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;It is part of the daily workflow.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Sashiko completes roughly 1200 reviews a day across over 100 kernel mailing lists. It sends review emails only to lists whose maintainers opt in, and already over 70 of them have done so. Developers also engage directly with the reports: More than 1500 kernel developers have replied on the mailing lists, and their pushback on false alarms has directly driven improvements to Sashiko's prompts and verification engine. In fact, &lt;/span&gt;&lt;a href="http://lwn.net" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;LWN.net&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; recently &lt;/span&gt;&lt;a href="https://lwn.net/Articles/1096963/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;wrote&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; that &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Sashiko has already become an important part of the kernel's development process&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; and was &lt;/span&gt;&lt;a href="https://lwn.net/Articles/1100241/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;the top bug reporter&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; of issues in the 7.2 release cycle.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;It prevents new bugs and uncovers old ones.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; When Sashiko flags a bug in a proposed patch, the author typically fixes it in the next revision before the change is ever merged. Many developers run Sashiko locally or in corporate private instances. The bugs that Sashiko helped to prevent leave no trace in the git commit log, making them hard to count directly. What &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;can&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; be counted are fixes for preexisting bugs that Sashiko finds and reports as a byproduct of reviewing new changes: Sashiko is cited in over 2000 mainline kernel commits and in over 500 of this year's official Linux kernel CVEs, spanning drivers, core kernel code, networking, GPUs, memory management, and virtualization.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Fewer bugs are slipping into mainline releases.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Across the 18 kernel releases prior to Sashiko (Linux 6.2 through 6.19, covering 254000 commits), an average of 3.4% of merged commits required a follow-up fix within 60 days. Since Sashiko rolled out, that 60-day fix rate has fallen to 2.8% in Linux 7.1 and 2.2% in Linux 7.2, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;a 35% drop&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; (with the 30-day fix rate falling from 2.3% to 1.4%). It’s too early to make strong claims, and obviously many other factors can influence release-to-release numbers, so we cannot solely credit Sashiko for the entire drop, but the trend is optimistic.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-pull_quote"&gt;&lt;div class="uni-pull-quote h-c-page"&gt;
  &lt;section class="h-c-grid"&gt;
    &lt;div class="uni-pull-quote__wrapper h-c-grid__col h-c-grid__col--8 h-c-grid__col-m--6 h-c-grid__col-l--6
      h-c-grid__col--offset-2 h-c-grid__col-m--offset-3 h-c-grid__col-l--offset-3"&gt;
      &lt;div class="uni-pull-quote__inner-wrapper h-c-copy h-c-copy"&gt;
        &lt;q class="uni-pull-quote__text"&gt;Since Sashiko was rolled out, the percentage of merged commits in the Linux kernel that received a follow-up fix in 60 days has dropped by 35%.&lt;/q&gt;

        
      &lt;/div&gt;
    &lt;/div&gt;
  &lt;/section&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;A clean review carries real signal.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; If Sashiko's warnings were mostly noise, commits merged with warnings would need follow-up fixes no more often than commits that passed cleanly. In practice, across 21,500 tracked mainline commits, patches that received a clean Sashiko review required a follow-up fix within 60 days just 0.9% of the time — less than a third of the historical baseline. Patches merged despite a critical Sashiko warning needed a fix 7.6% of the time — an 8x higher risk (and 6.3x higher even after adjusting for patch size). &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;That gap is what makes Sashiko useful to maintainers: A clean pass instills real confidence, while a critical warning is a strong reason to take another look at the proposed patch.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/2_IdMDvr5.max-1000x1000.png"
        
          alt="2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This all tracks with what Linus Torvalds wrote in a mailing-list discussion a few weeks after Sashiko’s launch:&lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;"As you say, Sashiko tends to be quite good. Even when it flags something unnecessarily and there's no actual problem, there's likely to be a reason for the question. And we've seen some truly stellar reports too. Things that are deep and subtle, and the AI is 100% correct. So anybody who thinks that all AI is slop is in denial about the current state."&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; - &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Linus Torvalds&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; (&lt;/span&gt;&lt;a href="https://lore.kernel.org/all/CAHk-=wi_drr4Ls9KtXW1k8L2FUDF0YdnyjvKmPgLXHDFnnRWEg@mail.gmail.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;on lore.kernel.org&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;)&lt;br/&gt;&lt;br/&gt;&lt;/span&gt;&lt;/p&gt;
&lt;div align="left"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;&lt;table style="width: 97.389%;"&gt;&lt;colgroup&gt;&lt;col style="width: 48.2659%;"/&gt;&lt;col style="width: 51.7341%;"/&gt;&lt;/colgroup&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th scope="col" style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Sashiko at a glance&lt;/strong&gt;&lt;/p&gt;
&lt;/th&gt;
&lt;th scope="col" style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Since March 2026&lt;/strong&gt;&lt;/p&gt;
&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Patch reviews completed&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;241,000 (~1200+ / day)&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Opted-in kernel subsystem lists&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;75+&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Mainline commits crediting Sashiko&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;2,000+&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;2026 Linux kernel CVEs crediting Sashiko&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;528&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;60-day post-merge fix rate&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Down from 3.4% baseline to 2.2% (Linux 7.2)&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;What we learned from building and running Sashiko&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;First, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;benchmark-driven development beats intuition.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; While developing Sashiko, we tried many ideas that looked very reasonable on their face but didn’t end up working out. Other ideas were less obvious but worked surprisingly well. What was crucial: Early on, we built a benchmark out of nearly 1,000 real Linux kernel bugs and used it to evaluate every major Sashiko change. Every bug in this set was a real mistake that had passed human review by kernel experts, had been merged into mainline Linux, and later had to be fixed. On that benchmark, Sashiko with the new &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Gemini 4 Carbon&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; model catches &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;75.8%&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; of the bugs that human reviewers missed (using high thinking level).&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Second, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;coding agents are not great at reviewing their own code.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Models are trained to produce a working solution within a limited token budget. Asking these agents to review their own code might work well for isolated tasks, but for real, large codebases, creating a functional solution is often just 5-10% of the work. What’s really hard is to produce a solution that works all of the time, is well-integrated into the surrounding code, and doesn’t introduce subtle regressions. This is often substantially more complex and requires additional effort. Similar to human developers, agents benefit from external code review: Reviewers often spot issues that the initial author missed, even if they are not more experienced.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Third, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;open source and shared community ownership and development help to build trust.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; By being fully open and transparent from the &lt;span style="vertical-align: baseline;"&gt;beginning&lt;/span&gt;, Sashiko allowed anyone to run it the way they wanted: using different models (including open source and local models), customizing prompts, or even adopting Sashiko to other-than-Linux projects. This helped with community adoption and brought over 100 contributors to improve Sashiko, including some prominent kernel developers.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;More and more engineers are putting local Sashiko instances into their coding-agent loops, so patches are checked against the community's review rules before a human maintainer ever sees them. We use the same workflow to build Sashiko itself ("Sashiko for Sashiko"), running it to review every change to its codebase and prompt pipeline both locally and through a Github integration.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Although Sashiko started in the Linux kernel, the need for an agentic trust layer applies across critical open-source software. Because the core review engine does not depend on C or on kernel-specific rules, the Linux kernel playbook (combining an open verification pipeline, community-owned domain guides, and a benchmark mined from historically known issues and fixes) can likely be applied to other foundational open-source projects such as GCC, LLVM, or Git.&lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;"Now, I think the other side of AI - and this is where Sashiko comes in - is that it basically does end up being a tool that can then increase the trust in the patch. So AI giveth, and AI taketh away. But the basic issue shouldn't be AI per se, it should be that notion of 'trust'." - &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Linus Torvalds&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;a href="https://lore.kernel.org/all/CAHk-=wgbGarE7Ozw4VG6oUKDj9pk-8DRoDiX00bo1MwEMm9UWQ@mail.gmail.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;on LKML&lt;/span&gt;&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;You can check out Sashiko at &lt;/span&gt;&lt;a href="https://sashiko.dev" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;sashiko.dev&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, or find the code on &lt;/span&gt;&lt;a href="https://github.com/sashiko-dev/sashiko" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;GitHub&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to run it in your own development workflow.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 18 Nov 2026 17:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/application-development/lessons-from-building-sashiko-an-agentic-code-review-system/</guid><category>AI &amp; Machine Learning</category><category>Open Source</category><category>Application Development</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Sashiko: Lessons from building an agentic code review system for Linux</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/application-development/lessons-from-building-sashiko-an-agentic-code-review-system/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Roman Gushchin</name><title>Software Engineer, AI &amp; Infrastructure</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Eric Brewer</name><title>VP, Infrastructure &amp; Google Fellow, AI &amp; Infrastructure</title><department></department><company></company></author></item><item><title>Don’t just prompt AI — architect it: Introducing the Professional Agentic Architect Certification</title><link>https://cloud.google.com/blog/topics/training-certifications/google-cloud-agentic-architect-certification-dont-just-prompt-ai-design-it/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;It just got easier to prove your expertise in building production-ready agents. Today, Google Cloud is introducing the &lt;/span&gt;&lt;a href="https://cloud.google.com/learn/certification/agentic-architect/"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Professional Agentic Architect certification&lt;/span&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;as the latest milestone in our &lt;/span&gt;&lt;a href="https://developers.google.com/program/gear" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Agent Ready (GEAR)&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; initiative. Nearly one million developers are using GEAR to skill up on agents, but what makes this particular certification so special is its adoption within Google’s walls. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Over the past four months, Google Cloud has trained thousands of forward deployed engineers and strategic partners on our own advanced curriculum. Originally built as an internal training initiative, we are now opening it up to the entire developer community and partner ecosystem through this new certification. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As the VP who manages our forward-deployed engineers here at Google Cloud (Ravi) and the managing director of Google Cloud learning (Gary), we’ve seen just how effective this program has been inside Google at helping our teams rethink the way they build, deploy, and use AI in everyday in almost every way. We’re equally about what this new certification can bring to everyone — developers, engineers, and anyone else ready to demonstrate their agentic leadership.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=AaQTFwEIy78"
      data-glue-modal-trigger="uni-modal-AaQTFwEIy78-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        

        &lt;div class="article-video__aspect-image"
          style="background-image: url(https://storage.googleapis.com/gweb-cloudblog-publish/images/PAAC_AnthemVideo_Thumbnail_A.max-1000x1000.png);"&gt;
          &lt;span class="h-u-visually-hidden"&gt;New Google Cloud Agentic Architect Certification&lt;/span&gt;
        &lt;/div&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-AaQTFwEIy78-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="AaQTFwEIy78"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=AaQTFwEIy78"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;A new hybrid format to validate your real-world skills&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This is not a traditional multiple choice exam. Google has introduced a new hybrid format that requires you to prove both conceptual know-how and hands-on execution.&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Part 1 (The multiple choice proctored exam): &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Evaluates your system design choices, architectural standards, cost optimization, and security governance.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Part 2 (Hands-on labs on Google Skills):&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Puts you in an active cloud environment to build, test, and debug live agentic workflows.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Why earning this certification can help advance your career&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As enterprises look to find top AI talent, this certification immediately signals that you have the advanced skills necessary to design secure, governed, and high-performance multi-agent systems. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href="https://www.ipsos.com/en-us/future-ai-and-technical-skill-building" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Our research with Ipsos on AI skill building&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; shows that certifications drive more recruiter outreach: &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;82% of learners using Google Skills say that mentioning Google Cloud certifications on their resume drives more recruiter responses&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;.&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;And it’s not just about getting the call — it’s about being able to answer it. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;That same research showed that &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;90% of cloud professionals say credentials like Google Cloud certifications give them the verified know-how needed to fill in-demand roles. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;What’s covered on the exam&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The certification validates your proficiency across five key domains:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Build agents using low-code tools (democratizing rapid agent prototyping).&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Use coding agents for application development (supercharging your coding productivity).&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Develop custom agents on the &lt;/span&gt;&lt;a href="https://cloud.google.com/products/gemini-enterprise-agent-platform"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini platform&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; with open source frameworks.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Evaluate and deploy agentic workflows (monitoring, optimizing, and ensuring performance).&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Secure and govern agentic workflows (protecting sensitive enterprise data and ensuring model safety).&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Battle-tested by thousands of Google Cloud engineers and partners — now available to you&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Enterprise demand for agentic workloads has skyrocketed. As our customers move past AI prompting and run their most business critical workloads on Google Cloud, we’ve built our own team that delivers the agentic future with them — at scale. Google Cloud partners and Google’s forward deployed engineers are embedded directly with customers around the world and are prepared to solve deep technical challenges because of this rigorous curriculum and certification. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;While building our forward-deployed engineering program, we wanted to standardize training and upskilling for customers, Google Cloud partners, and our own engineers with the Professional Agentic Architect certification. This ensures our own teams are using the same world-class standards that we recommend to our customers and partners, allowing us to learn from each other and sharpen our skills together. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Skill up for your certification at no cost today&lt;/strong&gt;&lt;/h3&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Register for &lt;/span&gt;&lt;a href="https://developers.google.com/program/gear" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Agent Ready&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to get access to 35 free monthly learning credits on Google Skills. For Google Cloud customers, &lt;/span&gt;&lt;a href="https://developers.google.com/program/gear/getcertified" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Get Certified&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; provides a cohort-based, guided certification journey at no cost. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;For Google Cloud partners, register for &lt;/span&gt;&lt;a href="https://rsvp.withgoogle.com/events/partner-learning/certprograms?pli=1" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Partner Certification Kickstart&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and review the &lt;/span&gt;&lt;a href="https://partner.skills.google/paths/4531" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;partner learning path&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Review the &lt;/span&gt;&lt;a href="https://services.google.com/fh/files/misc/professional_agentic_architect_exam_guide_english.pdf" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;exam guide&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;a href="https://www.skills.google/paths/4525" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;learning path&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://docs.google.com/forms/d/1fi1gqbWsNCFxm1CT9V-JNGff1lyqfU9kNWAaaQMvRi4/viewform?edit_requested=true" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;sample questions&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Get hands-on with Google Skills labs and preparation modules to get comfortable developing custom agents on Gemini platform with open source frameworks. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://cloud.google.com/learn/certification/agentic-architect/"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Register&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; for the multiple-choice exam, complete the hands-on lab requirements, and claim your digital badge to showcase your expertise to the world.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Don't just prompt AI — architect it. &lt;/span&gt;&lt;a href="https://cloud.google.com/learn/certification/agentic-architect/"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Get certified today&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 11 Nov 2026 17:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/training-certifications/google-cloud-agentic-architect-certification-dont-just-prompt-ai-design-it/</guid><category>AI &amp; Machine Learning</category><category>Application Development</category><category>Application Modernization</category><category>DevOps &amp; SRE</category><category>Training and Certifications</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/PAA_blog_header_approved.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Don’t just prompt AI — architect it: Introducing the Professional Agentic Architect Certification</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/PAA_blog_header_approved.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/training-certifications/google-cloud-agentic-architect-certification-dont-just-prompt-ai-design-it/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Gary Eimerman</name><title>Managing Director, Google Cloud Learning</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Ravi Rajamani</name><title>VP, Applied AI &amp; FDEs</title><department></department><company></company></author></item><item><title>TorchTPU is now open source: Run native PyTorch on TPUs</title><link>https://cloud.google.com/blog/topics/ai-infrastructure/torchtpu-is-now-open-source-run-native-pytorch-on-tpus/</link><description>&lt;div class="block-paragraph"&gt;&lt;p data-block-key="3zjpw"&gt;Today at PyTorch Conference 2026, we announced a range of new products and integrations designed to help you run PyTorch effectively on Google Cloud or in your own infrastructure. Notably, TorchTPU, our native PyTorch stack for Google TPUs, is now available as open source on GitHub. With TorchTPU, TPUs join GPUs and CPUs as a native option for PyTorch. You can now train, serve, and deploy models on TPUs using the tools you already rely on, including vLLM, SGLang, llm-d, TorchTitan, Ray, TorchRec, Hugging Face, LiteRT. Ultimately, this provides our customers choice across a broad range of workload-optimized computing options including TPUs, GPUs, and CPUs, with a consistent, open software and operational model across the full portfolio.&lt;/p&gt;&lt;h3 data-block-key="75cat"&gt;&lt;b&gt;Run PyTorch code natively on TPUs&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="2i98e"&gt;Co-developed with the PyTorch team at Meta, &lt;a href="https://github.com/google-pytorch/torch_tpu" target="_blank"&gt;TorchTPU&lt;/a&gt; lets you run existing PyTorch workloads on TPUs with minimal code changes, using the familiar APIs. TorchTPU lets you debug your models with eager mode, then transition to production-level performance via torch.compile to optimize TPU utilization and reduce memory overhead. By supporting standard PyTorch APIs, advanced diagnostics, and native distributed primitives out of the box, TorchTPU unlocks high performance and operational efficiency from day one.&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p style="padding-left: 40px;"&gt;&lt;span style="vertical-align: baseline;"&gt;"&lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;At Meta, advancing the frontier of AI requires immense compute scale and deep hardware diversity&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;We cannot be constrained by a single hardware architecture. TorchTPU is a massive step forward — allowing us to co-develop hardware flexibly with the core machine learning systems team while supporting major LLM, Ads, and Recommendation systems workloads on key production use cases. It enables us to deploy our PyTorch workloads to Google TPUs, provide direct feedback to the engineering teams, and build the ecosystem flexibility everyone needs.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;"- Jana van Greunen, Director of PyTorch Engineering, Meta&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-paragraph"&gt;&lt;p data-block-key="3zjpw"&gt;We also partnered with Hugging Face so that popular open-weight models across the Hugging Face Hub run natively on TorchTPU. In Transformers v5, deploying models to TPUs is now as simple as passing &lt;code&gt;device_map="tpu"&lt;/code&gt;. The Diffusers library runs natively on TPUs with support for eager mode, torch.compile, and tensor parallelism to scale large generative models across multiple accelerators. To maintain consistent reliability, we also built dedicated CI/CD pipelines for continuous nightly testing of the Hugging Face Transformers suite on TPU infrastructure.&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p style="padding-left: 40px;"&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;“Hugging Face is committed to being an open, neutral platform where AI builders can choose the models, frameworks and hardware that work best for them. Hardware diversity is essential to making open-source AI accessible to everyone. Our collaboration with Google to bring native TorchTPU support to our core libraries, including Transformers, puts that commitment into practice. Together, we’re making it easier for the community to run popular models from the Hub efficiently on Google Cloud TPUs using familiar PyTorch workflows. Giving developers more choice in how and where they build helps the entire open-source ecosystem grow.”&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; - Jeff Boudier, VP Product, Hugging Face&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=u4n558Ul6w0"
      data-glue-modal-trigger="uni-modal-u4n558Ul6w0-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        &lt;img src="//img.youtube.com/vi/u4n558Ul6w0/maxresdefault.jpg"
             alt="Watch Introducing TorchTPU: Native PyTorch on Google Cloud TPUs to learn more"/&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-u4n558Ul6w0-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="u4n558Ul6w0"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=u4n558Ul6w0"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph"&gt;&lt;h3 data-block-key="94fb3"&gt;&lt;b&gt;Optimizing inference&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="emr2n"&gt;As reasoning and agentic models scale, inference cost and latency become the primary bottlenecks in production. We’ve integrated TorchTPU into widely used open-source serving engines and paired them with Kubernetes-native fleet routing across TPUs and GPUs to help you optimize throughput per dollar.&lt;/p&gt;&lt;p data-block-key="74rn3"&gt;This foundation allows developers to scale workloads across diverse hardware while minimizing the friction of platform-specific tuning. To bring these framework-level enhancements to life in production, we have collaborated with leading open-source communities and industry partners. In the sections below, we explore how these partnerships are driving breakthrough efficiency across high-throughput serving with vLLM, structured agentic execution with SGLang, and large-scale distributed fleet orchestration with llm-d.&lt;/p&gt;&lt;p data-block-key="98pg6"&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;div data-draftjs-conductor-fragment='{"blocks":[{"key":"37eh","text":"vLLM: High-throughput serving and memory efficiency","type":"unstyled","depth":0,"inlineStyleRanges":[],"entityRanges":[],"data":{}},{"key":"gq3","text":"Developed in partnership with Inferact, the native PyTorch vLLM backend on TorchTPU enables developers to deploy open-weight models directly on TPUs with optimized, production-ready serving features and custom Pallas kernels. This integration supports continuous batching, quantization, and multi-chip parallelism on TPU hardware, while preserving upstream worker models, schedulers, and API servers to improve cost-performance ratios, without requiring changes to existing inference stacks.","type":"unstyled","depth":0,"inlineStyleRanges":[],"entityRanges":[{"offset":59,"length":4,"key":0}],"data":{}}],"entityMap":{"0":{"type":"LINK","mutability":"MUTABLE","data":{"url":"https://github.com/vllm-project/vllm-torchtpu"}}}}'&gt;
&lt;div class="Draftail-block--unstyled" data-block="true" data-editor="7vaaq" data-offset-key="98pg6-0-0"&gt;
&lt;h4 class="public-DraftStyleDefault-block public-DraftStyleDefault-ltr" data-offset-key="98pg6-0-0"&gt;&lt;span style="text-decoration: underline;"&gt;vLLM: High-throughput serving and memory efficiency&lt;/span&gt;&lt;/h4&gt;
&lt;/div&gt;
&lt;div class="Draftail-block--unstyled" data-block="true" data-editor="7vaaq" data-offset-key="2sq9l-0-0"&gt;
&lt;div class="public-DraftStyleDefault-block public-DraftStyleDefault-ltr" data-offset-key="2sq9l-0-0"&gt;&lt;span data-offset-key="2sq9l-0-0"&gt;Developed in partnership with Inferact, the native PyTorch &lt;/span&gt;&lt;a class="TooltipEntity" data-draftail-trigger="true" href="https://github.com/vllm-project/vllm-torchtpu" rel="noopener" role="button" target="_blank"&gt;&lt;span data-offset-key="2sq9l-1-0"&gt;vLLM&lt;/span&gt;&lt;/a&gt;&lt;span data-offset-key="2sq9l-2-0"&gt; backend on TorchTPU enables developers to deploy open-weight models directly on TPUs with optimized, production-ready serving features and custom Pallas kernels. This integration supports continuous batching, quantization, and multi-chip parallelism on TPU hardware, while preserving upstream worker models, schedulers, and API servers to improve cost-performance ratios, without requiring changes to existing inference stacks.&lt;/span&gt;&lt;/div&gt;
&lt;div class="public-DraftStyleDefault-block public-DraftStyleDefault-ltr" data-offset-key="2sq9l-0-0"&gt; &lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="vertical-align: baseline;"&gt;"&lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Our mission at Inferact has always been to make high-performance inference accessible, reliable, and cost-effective across the entire AI ecosystem. We are thrilled to partner on the launch of TorchTPU and bring first-class support for vLLM to TPUs. By bridging PyTorch’s optimized TPU capabilities directly into vLLM, developers can immediately run production workloads on TPUs with the high throughput, low latency, and memory efficiency they expect from our stack.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;" - Simon Mo, CEO and co-founder, Inferact&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h4&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;SGLang: Accelerating agentic and structured workloads&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Developed in partnership with RadixArk, our native PyTorch backend for &lt;/span&gt;&lt;a href="https://github.com/sgl-project/sglang-torchtpu/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;SGLang&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; accelerates highly structured and agentic workloads on TPUs. SGLang uses RadixAttention to reuse KV cache across requests, allowing repeated prompt prefixes to skip recomputation entirely. SGLang-TorchTPU preserves upstream worker models, schedulers, and API servers, using a disaggregation transfer backend that also acts as the I/O backend for HiCache to optimize latency and throughput for complex multi-step generation.&lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="vertical-align: baseline;"&gt;"&lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;At RadixArk, TorchTPU has supercharged our development velocity as we roll out SGLang support across diverse silicon. It’s given us a painless path to native TPU acceleration, passing massive efficiency gains and flexible compute choices directly down to our customers.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;" - Banghua Zhu, CTO and co-founder, RadixArk&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;llm-d for distributed LLM inference&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To run inference engines across large fleets, we are releasing llm-d 1.0 with Red Hat, IBM, and a community of more than 500 contributors and 25 organizations. As a Cloud Native Computing Foundation (CNCF) Sandbox project, llm-d connects two open-source communities: PyTorch, where vLLM and SGLang serve models, and Kubernetes, where distributed systems run and scale in production. The llm-d router sends each request to the replica that already holds its context, using prefix-cache-aware routing. It also splits compute-heavy prefill from memory-heavy decode. llm-d 1.0 makes that ready for production with multi-tenant flow control, router high availability, stable v1 APIs, and session-aware routing for agents. On Google Kubernetes Engine (GKE), you can have the choice of running on either GPUs or Google TPUs, with a common llm-d endpoint, and choose the hardware that fits each workload.&lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="vertical-align: baseline;"&gt;"&lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;At Zenteiq, we use GKE Inference Gateway powered by llm-d to accelerate synthetic data generation for our sovereign Scientific Foundation Models. By leveraging load-aware and prefix-cache-aware routing, we have achieved over 4× higher aggregate inference throughput, increasing daily generated token capacity from 1.47 billion to 5.97 billion tokens per 1,000 TPU v6e accelerators. The open-source foundation of llm-d, integrated with GKE, enables us to efficiently serve multiple model families at scale without developing and maintaining a custom inference scheduler.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;" - Prof. Sashikumaar Ganesan, Founder, Zenteiq.ai&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Training at scale&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Scaling PyTorch training across thousands of accelerators requires fault-tolerant, compile-aware distributed frameworks. We have integrated TorchTPU into the PyTorch community’s core training, reinforcement learning (RL), and orchestration frameworks.&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;TorchTitan: Distributed high performance training at scale&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;a href="https://github.com/google-pytorch/torchtitan" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;TorchTitan&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; is the PyTorch-native reference implementation for scalable distributed training and features a production-grade integration with TorchTPU. By connecting to PyTorch's &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;torch.compile&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; and TorchTitan’s compile friendly SimpleFSDP, specialized Pallas kernels, and graph-level optimizations, developers can build and train models on TPUs with high efficiency at scale.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;TorchTitan provides recipes for pre-training large dense models (such as Llama 3 70B/405B, Qwen 3 8B/32B, and MuseGlimmer 30B), Mixture-of-Experts (MoE) such as Qwen 3 235B, and diffusion transformers (such as FLUX.1 12B), and has been validated running with up to 8000 Ironwood TPUs (TPU v7x). Beyond initial pre-training, TorchTitan serves as the scalable trainer inside post-training pipelines like VeRL for Group Relative Policy Optimization (GRPO) and RL alignment.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;To test large-scale efficiency on demanding open architectures, TorchTPU’s training paths via TorchTitan demonstrated up to 52% Model Flops Utilization (MFU) on Qwen 3 8B, showing near-linear scaling and stability across up to 8,192 Ironwood TPUs.&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;VeRL: Scalable post-training and reinforcement learning in preview&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;RL is a foundational technique for frontier labs that are advancing model capabilities, as well as for enterprises customizing models for domain-specific tasks. Yet, orchestrating distributed RL workloads remains one of the most complex challenges in modern AI infrastructure. Bridging that gap, &lt;/span&gt;&lt;a href="https://github.com/verl-project/verl/tree/main/examples/tpu/grpo" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;VeRL&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; has quickly become one of the most popular RL frameworks, supporting everything from verifiable math and coding rewards, to multi-turn agentic tool use at massive scale. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To help expand hardware choice for this thriving community, we upstreamed a TPU backend and an end-to-end reference recipe to VeRL. Built on a disaggregated trainer-sampler architecture designed to maximize training efficiency, our GRPO reference recipe on mathematical reasoning datasets demonstrates how developers can deploy VeRL workloads on TPUs while preserving their existing, hardware-agnostic model and algorithm code. &lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Ray: Orchestrating heterogenous compute&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We’ve also added native TorchTPU support to &lt;/span&gt;&lt;a href="https://github.com/ray-project/ray" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Ray&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, enabling you to orchestrate PyTorch data processing, training, RL, and serving across CPUs, GPUs, and TPUs within a single workflow. With a few lines of resource configuration, you can schedule each stage of a heterogeneous AI pipeline onto the accelerator best suited for the job.&lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="vertical-align: baseline;"&gt;“&lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;At Ray Summit we talked about the PARK stack, PyTorch, AI models, Ray, and Kubernetes, because that's what teams are building production AI on today. TorchTPU embeds Google TPUs directly in that industry-standard stack. We've also worked closely with the Google team to support TPUs as a first-class accelerator in Ray. Coupled with native PyTorch support, teams can now seamlessly develop, scale, and monitor their workloads on TPUs.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;" - Edward Oakes, Chief Architect, Anyscale&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Deployment flexibility and operational efficiency&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;True portability requires standardizing deployment tools across different hardware profiles, including edge devices and cloud accelerators.&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;LiteRT joins the PyTorch Ecosystem&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Many PyTorch models need to run on phones, laptops, browsers, and embedded devices. Today, &lt;/span&gt;&lt;a href="https://github.com/google-ai-edge/litert" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;LiteRT&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (formerly TensorFlow Lite), Google's high-performance on-device AI inference, joins the PyTorch ecosystem. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With &lt;/span&gt;&lt;a href="https://github.com/google-ai-edge/litert-torch" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;LiteRT-Torch&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, you convert a PyTorch model with a single &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;litert-torch.convert()&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; call. It builds on&lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt; torch.export()&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; and Core ATen operators, so your model code stays in PyTorch. For LLMs, LiteRT Torch exports Hugging Face Transformer models to run with &lt;/span&gt;&lt;a href="https://github.com/google-ai-edge/LiteRT-LM" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;LiteRT-LM&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. The converted models run on Android, iOS, macOS, Windows, Linux, web, and IoT devices, with hardware acceleration across CPUs (via XNNPack), GPUs (via open source &lt;/span&gt;&lt;a href="https://developers.googleblog.com/ml-drift-next-gen-gpu-aiml-inference-at-the-edge/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;ML Drift&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;), and NPUs from major chipsets provided through a single API. &lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;TPU agent suite in early preview&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To expand Pytorch adoption on TPUs, we are announcing that the &lt;/span&gt;&lt;a href="https://github.com/google-pytorch/tpu-agent-suite" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;TPU agent suite&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; is in preview on GitHub. Integrating through plug-ins with several coding harness-model combinations, the TPU agent suite helps automate onboarding, diagnostics, user questions, and inference performance tuning for many PyTorch models. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=hK3xqW_HrP0"
      data-glue-modal-trigger="uni-modal-hK3xqW_HrP0-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        &lt;img src="//img.youtube.com/vi/hK3xqW_HrP0/maxresdefault.jpg"
             alt="Watch Introducing the TPU agent suite (preview): Open-source agents for PyTorch on Cloud TPU to learn more"/&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-hK3xqW_HrP0-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="hK3xqW_HrP0"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=hK3xqW_HrP0"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph"&gt;&lt;h3 data-block-key="j085k"&gt;&lt;b&gt;Get started with PyTorch on Google Cloud&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="9hspq"&gt;Together, these releases reinforce our commitment to making the Google TPU software ecosystem an open, developer-friendly environment for the PyTorch community. By bringing native TPU support upstream into the PyTorch stack across TorchTPU, Hugging Face, vLLM, SGLang, llm-d, TorchTitan, VeRL, and Ray — and contributing LiteRT to the PyTorch Ecosystem — we are giving you the flexibility to build, adapt, and scale AI on your terms.&lt;/p&gt;&lt;p data-block-key="5qbfp"&gt;Visit the &lt;a href="https://cloud.google.com/tpu/docs/pytorch-on-tpu"&gt;Google Cloud documentation&lt;/a&gt; to get started running PyTorch on TPUs today.&lt;/p&gt;&lt;/div&gt;</description><pubDate>Tue, 20 Oct 2026 11:30:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/ai-infrastructure/torchtpu-is-now-open-source-run-native-pytorch-on-tpus/</guid><category>TPUs</category><category>Application Development</category><category>Open Source</category><category>AI infrastructure</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>TorchTPU is now open source: Run native PyTorch on TPUs</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/ai-infrastructure/torchtpu-is-now-open-source-run-native-pytorch-on-tpus/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Mark Lohmeyer</name><title>VP &amp; GM, AI &amp; Computing Infrastructure</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Bill Jia</name><title>Vice President, Engineering</title><department></department><company></company></author></item><item><title>Temporal Serverless Workers on Cloud Run: Scale workers ahead of demand</title><link>https://cloud.google.com/blog/products/serverless/temporal-serverless-workers-on-cloud-run-worker-pools/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The open-source &lt;/span&gt;&lt;a href="https://temporal.io/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Temporal&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; execution platform is rapidly becoming a foundational layer of modern software infrastructure — especially for orchestration in distributed systems and AI agent workloads. However, deploying and scaling workers in Temporal comes with unique challenges. Workers poll for tasks instead of serving requests, so request-driven serverless platforms can be an awkward fit. This often results in running worker fleets on Kubernetes or VMs, which you have to manage and pay for continuously. You build a metrics pipeline and tune autoscaling triggers, and scaling still trails bursts because of metrics lag. Every deployment means juggling upgrades and versions, and you pay for that compute whether or not there's work in the queue.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We built &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/run/docs/deploy-worker-pools"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Run worker pools&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; for exactly this kind of pull-based work: Your instances connect to a queue or stream, take the next item, and process it. Starting today, &lt;/span&gt;&lt;a href="https://temporal.io/cloud" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Temporal Cloud&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, which is the fully managed version of Temporal’s open-source platform, can scale them for you. Currently in preview, &lt;/span&gt;&lt;a href="https://docs.temporal.io/serverless-workers" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Temporal Serverless Workers&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; on Cloud Run let you deploy your workers as a worker pool, and Temporal sizes the pool to match the work on your task queues. When there's no work, it can scale the pool to zero. There's no cluster or autoscaler to run, and your worker code stays the same.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;How Cloud Run worker pools fit with Temporal&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;A worker pool is Cloud Run's resource for continuous background work. It has no public endpoint, and each instance is allocated CPU for its whole lifetime, which matches how a Temporal worker runs. Temporal manages your Cloud Run worker pools as follows:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Your Temporal workers run as Cloud Run worker pool instances in your project and poll Temporal Cloud for tasks.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Temporal Cloud tracks your task queues and sets the worker pool's instance count through the Cloud Run Admin API, using the IAM permissions you grant. Temporal never opens a connection to your instances, so you don't need to open ports or add firewall rules for it.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;You run each build version as its own Cloud Run worker pool. An older build's worker pool can sit at zero instances until a workflow that started on it needs it again.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Your Temporal workers run in your project, under your service account, and on your network. Temporal decides how many of them run.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;What you can build with Temporal Serverless Workers&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The combination of Cloud Run worker pools with Temporal Serverless Workers opens up a variety of interesting use cases:&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Autoscaled agentic workers:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; AI agents run multi-step chains of LLM reasoning and tool calls. Temporal's durable execution and Cloud Run worker pools work together here. Temporal workers spin up instantly to handle bursts of LLM steps, embeddings or API calls, then scale to zero when the queue empties. Workers get keyless access to Google Cloud AI APIs. You can also attach a GPU for inference, and the workers attached to the GPUs will still scale to zero. Temporal runs each model and tool call as a durable activity; if a worker disappears mid-step, the activity retries on another worker and the workflow continues.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Long-running &lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt;Workflows&lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt;:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;a href="https://docs.temporal.io/workflows" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Temporal Workflows&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; can run indefinitely, allowing for building applications that track the state of real-world resources over long periods of time. This pattern is typically known as the “&lt;/span&gt;&lt;a href="https://docs.temporal.io/design-patterns/entity-workflow%5C" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Entity Pattern&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;,” because the Workflow represents the state of some entity. However, these long-running Workflows don’t usually have consistent compute requirements: Some individual steps are short, some are long, and other times, the Workflow is sleeping, waiting for external input, such as from a human or another system. Cloud Run worker pools provide flexible compute to support this pattern: Individual steps can run for as long as is required, and Temporal will automatically scale down the worker pool when there is no work to be done.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Bursty batch work:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Worker pools bill only while instances run, so a worker pool that sits idle between batch runs costs nothing while it's at zero instances. For jobs like billing runs or media transcoding, Temporal scales the pool out as batches arrive and back in once the queue drains. Your maximum instance count caps how many instances you pay for. Tasks beyond that capacity wait in the queue rather than being dropped.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Why Cloud Run worker pools for Temporal workers&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Before, running Temporal workers on Cloud Run services took some workarounds:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;an HTTP listener, because services expect containers to listen on a port&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;instance-based billing, so workers kept their CPU between requests&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;minimum instances, so the service didn't scale to zero&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By running Temporal workers on Cloud Run worker pools, you don’t need those workarounds anymore, and you have:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;No endpoint to manage.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; There's no public URL or load balancer to lock down, and no health-check port to implement.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Steady CPU at a lower price.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Instances don't idle or throttle. Cloud Run doesn't remove instances because they look idle, so Temporal alone decides when the pool shrinks. Worker pools list at almost 40% less per vCPU-hour and GB-hour than instance-billed Cloud Run services.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Scaling without revision churn.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Changing the instance count doesn't create a new revision, so Temporal can resize the pool as often as the load changes.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;How Temporal manages scaling and versioning of worker pools&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Temporal &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;decides when to add capacity using one main signal: the delay between a task landing in the task queue backlog and a worker picking it up. If that delay grows, the available workers are saturated. Scaling occurs as follows:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Server-side detection:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; The Temporal server tracks this backlog-to-worker latency for each task queue.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Push notification:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; When the latency crosses its trigger conditions, the server sends a push notification to the worker controller instance.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Provisioning:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; The worker controller then adds capacity to the worker pool. New instances take time to start and connect before they can poll for tasks, so capacity lags behind the signal.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Because Temporal keeps spare capacity, the pool grows ahead of demand instead of after tasks pile up. Scale-in is more conservative. Temporal waits for a stabilization period after the last time a task found no available worker (90 seconds by default), and it pauses between successive reductions. Cloud Run’s configurable graceful shutdown extends shutdown from 10 seconds to 10 minutes. That gives SIGTERM-aware workers time to finish or checkpoint. The instance count stays between a minimum and a maximum that you set. If the first task after an idle period can't wait for an instance to start, set the minimum above zero.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_Tu66Q25.max-1000x1000.png"
        
          alt="1"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Deploying new workflow code without breaking running executions is hard. Temporal workflows are stateful and deterministic, so if you swap their code mid-run, you get non-determinism errors. Serverless Workers on Cloud Run solve this by requiring Temporal Worker Versioning. You don't patch code with conditional branches. Instead, you deploy each Worker Deployment Version as its own worker pool deployment, identified by a &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;deploymentName&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; and a &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;buildId&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Using distinct worker pools solves two problems:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Workflows that started on a particular Worker Deployment Version continue to run and complete on that version.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Each Worker Deployment Version is scaled and drained individually, matching compute spend and utilization to the requirements of your application. As existing workflows complete and drain on a previous Worker Deployment Version, the underlying Worker Pool is eventually scaled down to zero automatically.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;You set a versioning behavior for each Workflow type, or set a Worker-level default:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;PINNED:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; A workflow stays on the Worker version (Build ID) it started on and finishes there. Best for short-running workflows, order processing and multi-step APIs.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;AUTO_UPGRADE:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; A workflow moves to the new "Current" or "Target" Worker Deployment Version as you roll it out. Best for long-running workflows, such as subscription billing or tasks that run for weeks.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Learn more about configuring worker versioning&lt;/span&gt; &lt;a href="https://docs.temporal.io/worker-versioning" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;here&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Get started&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Setup takes three steps:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Containerize your Temporal Worker and deploy the container as a Cloud Run worker pool, with the Build ID in the pool name. Keep credentials such as your Temporal API key in Secret Manager.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Connect Temporal to your worker pool by granting Temporal Cloud permission to manage the worker pool so it can scale it.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Create a Worker Deployment Version in Temporal that points at the pool's project, region, and name.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/2_GCHxzit.gif"
        
          alt="2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Temporal Serverless Workers on Cloud Run is in preview, so APIs and configuration may change before the stable release. To get started,&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Learn how&lt;/span&gt; &lt;a href="https://docs.temporal.io/serverless-workers/cloud-run" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Serverless Workers on Cloud Run&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; scale and handle versions.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Deploy your first worker with the&lt;/span&gt; &lt;a href="https://docs.temporal.io/production-deployment/worker-deployments/serverless-workers/cloud-run" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;setup guide&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;See&lt;/span&gt; &lt;a href="https://cloud.google.com/run/docs/deploy-worker-pools"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Run worker pools&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; for CPU, memory, GPU and networking options.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Sign up for the configurable graceful shutdown &lt;/span&gt;&lt;a href="https://forms.gle/ctAU6WDUVwJajx1D7" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;preview&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;</description><pubDate>Thu, 15 Oct 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/serverless/temporal-serverless-workers-on-cloud-run-worker-pools/</guid><category>Application Development</category><category>Cloud Run</category><category>AI infrastructure</category><category>Serverless</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Temporal Serverless Workers on Cloud Run: Scale workers ahead of demand</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/serverless/temporal-serverless-workers-on-cloud-run-worker-pools/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Sagar Randive</name><title>Product Manager</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Aniruddh Chaturvedi</name><title>Engineering Manager</title><department></department><company></company></author></item><item><title>Google is a Leader in the 2026 Gartner® Magic Quadrant™ for Cloud-Native Application Platforms</title><link>https://cloud.google.com/blog/products/application-development/2026-gartner-mq-for-cloud-native-application-platforms/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We are thrilled to announce that Google has been recognized as &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;a Leader&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; for the third year in a row in the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;2026 Gartner® Magic Quadrant™ for Cloud-Native Application Platforms (CNAP)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. We believe this placement in the Leaders quadrant validates our commitment to providing an accessible, developer-centric platform that accelerates onboarding and supports rapid prototyping across modern workloads.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/Figure_1_Magic_Quadrant_for_CloudNative_Ap.max-1000x1000.png"
        
          alt="Figure_1_Magic_Quadrant_for_CloudNative_Application_Platforms"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Our vision for an application-centric cloud focuses on enabling developers to prioritize writing code and building agents or traditional apps by removing infrastructure complexity. Google Cloud provides a unified execution environment supporting &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;serverless, containerized, and agentic deployment o&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;ptions. We believe our placement highlights Google's unique readiness to power both standard enterprise microservices and the next generation of autonomous AI applications.  &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Some key features and capabilities of our platform are highlighted below. &lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;strong style="vertical-align: baseline;"&gt;From idea to implementation&lt;/strong&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Generative AI has ushered in a wave of vibe coding, allowing anyone to go from an idea to a deployed application in a fraction of the time it used to take. To make this even smoother, Google Cloud integrates its serverless infrastructure with AI vibe-coding and prototyping tools. We also simplify access to Google Cloud resources with tools like &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/mcp/overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;managed MCP servers&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="http://github.com/google/skills" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;agentic skills&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; — packaged sets of instructions, scripts, and resources to teach an AI how to complete specialized, multi-step workflow. &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;One-click prototyping in Google AI Studio: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Developers can build and deploy &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/databases/vibe-coded-ai-studio-apps-with-firestore-firebase-cloud-sql?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;full-stack applications&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; directly within Google AI Studio, making it a great environment for prototyping and experimentation. With a single click, you can instantly package and publish your vibe-coded applications to &lt;/span&gt;&lt;a href="https://cloud.google.com/run?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Run&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Google-managed MCP servers:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; To enable AI agents to interact with cloud resources, we support official, fully managed &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/mcp/supported-products"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;remote MCP servers&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. An example is the Cloud Run MCP server (via &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;run.googleapis.com/mcp&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;), which allows developers to easily launch endpoints and deploy server-side logic. The MCP tools are deployed with a simple config, skipping cloud builds to launch code in seconds, saving valuable developer time. These fully managed servers are integrated with IAM and VPC Service Controls, and they leverage Model Armor for content security.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Google's official Skills Repository&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Level up your agents with additional, condensed expertise on various Google Cloud technologies. Published and available in Agent Registry, the repository includes &lt;/span&gt;&lt;a href="https://github.com/google/skills/tree/main/skills/cloud/cloud-run-basics" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;skills for Cloud Run&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, the Well-Architected Pillar (security, reliability, and cost optimization), and more.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Ready to try vibe coding yourself? &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Get hands on with this codelab to &lt;/span&gt;&lt;a href="https://codelabs.developers.google.com/deploy-from-aistudio-to-run#0" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;build a vibe-coded app and deploy it to Cloud Run&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;strong style="vertical-align: baseline;"&gt;From implementation to enterprise-ready&lt;/strong&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Translating prototypes into production-grade, secure, and cost-effective enterprise software is where Google Cloud excels, with a full suite of developer, architect, and platform engineering tools. From designing your application to optimizing day 2 operations, we offer the services and tools to help you &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;build&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;operate&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;, and &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;deploy&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; applications across their entire lifecycle, and you have the freedom to build with any language, any library, and any framework.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Build&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Build with &lt;/strong&gt;&lt;a href="https://antigravity.google/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Google Antigravity&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;At Google, we’re simplifying and expanding our development ecosystem behind the Antigravity harness, collapsing developer silos into a unified orchestration layer. By integrating multi-step AI reasoning directly into the developer workflow, Antigravity natively brings local codebase development to our cloud-native application platforms (e.g., Cloud Run).&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Design and deploy with &lt;/strong&gt;&lt;a href="https://docs.cloud.google.com/application-design-center/docs/overview"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Application Design Center&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt; (ADC): &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Now, you can bridge the gap between developer velocity and enterprise control, using Application Design Center to eliminate manual Terraform and YAML configuration. This platform engineering component helps teams design, standardize, and deploy template-driven applications on Google Cloud. It is also integrated as part of Gemini Cloud Assist design agent and published as an MCP server. With ADC, you can visually design your architecture using Cloud Run services, databases, and event brokers backed by automated Gemini Cloud Assist security templates. Beyond human-guided design, ADC enables programmatic orchestration at the time of no HITL (Human-in-the-Loop), allowing automated pipelines to provision policy-governed Terraform configurations directly and autonomously.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Operate &lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Intelligent investigations: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Integrating &lt;/span&gt;&lt;a href="https://cloud.google.com/products/gemini/cloud-assist?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Cloud Assist&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; with native telemetry creates an AI-driven framework for Day-2 incidents. When alerts fire, operators engage Gemini Cloud Assist to instantly synthesize logs and metrics, pinpoint root causes, and generate remediations — context that can be handed off to accelerate support escalations. Crucially, IAM permissions strictly govern all AI recommendations, and help to ensure explicit human-in-the-loop approval are required before any infrastructure changes occur.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Cost analysis and optimizations: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Machine learning algorithms learn natural seasonal traffic cycles to detect cost anomalies within minutes, triggering notifications to protect your bottom line without risking destructive infrastructure shutdown.  &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Deploy&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Reliability and high availability: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Cloud Run is a regional service by default, but you can deploy an app to multiple regions via a single gcloud command. Integrated with &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/run/docs/configuring/configure-service-health"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;service health&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, Cloud Run automates cross-region failover and failback. If a service in one region becomes unhealthy, traffic is automatically routed to the next-closest healthy region, failing back once the issue is resolved.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;An open platform: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;As a long-time and top contributor to the Cloud Native Computing Foundation (&lt;/span&gt;&lt;a href="https://www.cncf.io/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;CNCF&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;), we operate with an open-source-first strategy. By integrating foundational, community-driven technologies, we help enable application portability for enterprise customers who are increasingly demanding multi-cloud flexibility&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Ready to start deploying your apps to Google Cloud? Get hands on with these codelabs:  &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://codelabs.developers.google.com/build-and-deploy-gcp-with-antigravity#0" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Build an app with Antigravity on Cloud Run&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://codelabs.developers.google.com/next26/adc-apps#0" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Deploy a Cloud Run app via Application Design Center&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://codelabs.developers.google.com/next26/gemini-cloud-assist-cost-optimization#0" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Optimize application costs with Gemini Cloud Assist&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h2&gt;&lt;strong style="vertical-align: baseline;"&gt;From enterprise-ready to autonomous&lt;/strong&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;AI agents are software’s next frontier. They offer more than just increased productivity and efficiency; they can unlock exponential growth. To provide enterprises with robust agentic deployment options, Google Cloud provides a dedicated infrastructure stack tailored specifically to host, govern, and secure autonomous agent fleets. This stack seamlessly integrates with our Agent Development Kit (&lt;/span&gt;&lt;a href="https://adk.dev/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;ADK&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;) as well as other leading agentic frameworks to give developers maximum flexibility.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Gemini Enterprise Agent Runtime&lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;At the core of this stack is Gemini Enterprise Agent Platform and its dedicated &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/build/runtime"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Runtime&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, which delivers the serverless and containerized deployment options you need for enterprise-scale agent development, including the following capabilities:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Native personalization:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Built-in sessions and memory banks manage context and long-term state, preventing costs from ballooning.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent observability and tracing:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Built on OpenTelemetry (OTel) standards and &lt;/span&gt;&lt;a href="https://opentelemetry.io/blog/2026/genai-observability/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;agentic schemas&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, turnkey dashboards feature agent topology graphs and interactive trace logs that detail sessions, tool calls, and reasoning paths.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent evaluation and simulation:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Automated simulation tools allow developers to test agents against golden sets with side-by-side comparisons and simulate thousands of interactions to test edge cases.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Hosting agents on Cloud Run&lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;For customers requiring additional flexibility, granular control, or specific regulatory compliance, Cloud Run serves as an excellent serverless alternative to host your agents. Some of its latest features include:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Cloud Run instances &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;(coming soon)&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;This primitive manages individual, addressable, long-running singleton resources with integrated Cloud Storage volume mounts, allowing persistent background agents to be deployed cost-effectively.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Cloud Run &lt;/strong&gt;&lt;a href="https://cloud.google.com/blog/topics/developers-practitioners/google-cloud-run-sandboxes-are-in-public-preview?e=48754805"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;sandboxes&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Hard-isolated environments spin up in under 500 milliseconds to safely execute untrusted, model-generated code, protecting the host system from unauthorized access.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent security, governance, and auditability&lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;To securely deploy AI agents and prevent unmanaged shadow AI, enterprises need an ironclad governance framework. Google Cloud delivers this through &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/govern/agent-identity-overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Identity&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (non-human IAM with cryptographic IDs) to provide an auditable trail of all actions and reasoning; a centralized &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/agent-registry/overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Registry&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to manage approved agents, skills, tools and application artifacts,  and prevent unauthorized tool integrations; and an &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/govern/gateways/agent-gateway-overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Gateway&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to proxy traffic, enforce Model Armor policies, and actively block destructive actions. These features are available on &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/build/runtime"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Runtime&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; today and will be available soon on Cloud Run and Google Kubernetes Engine (GKE).&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Ready to start deploying agents? Check out various codelabs featuring Gemini Enterprise Agent Platform &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/13-demos-on-gemini-enterprise-agent-platform"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;here&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;strong style="vertical-align: baseline;"&gt;Build the future of cloud-native applications&lt;/strong&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Whether you’re a vibe coder deploying your first full-stack application, a software architect standardizing production microservices, or an enterprise team scaling a fleet of secure AI agents, Google Cloud delivers the simplicity, elasticity, and security you need. &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Read the full report:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Download your complimentary copy of the&lt;/span&gt; &lt;a href="https://cloud.google.com/resources/content/gartner-cloud-native-application-platforms-magic-quadrant-report-2026"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;2026 Gartner® Magic Quadrant™ for Cloud-Native Application Platforms (CNAP)&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;sub&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Magic Quadrant for Cloud-Native Application Platforms, By Mukul Saha, Alex Coqueiro, Prasanna Lakshmi Narasimha, Richard Watson, 3 August 2026&lt;/span&gt;&lt;/sub&gt;&lt;/p&gt;
&lt;p&gt;&lt;sub&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Gartner and Magic Quadrant are trademarks of Gartner, Inc. and/or its affiliates. This graphic was published by Gartner, Inc. as part of a larger research document and should be evaluated in the context of the entire document. The Gartner document is available upon request from Google. &lt;/span&gt;&lt;/sub&gt;&lt;/p&gt;
&lt;p&gt;&lt;sub&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Gartner does not endorse any company, vendor, product or service depicted in its publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner publications consist of the opinions of Gartner’s business and technology insights organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this publication, including any warranties of merchantability or fitness for a particular purpose.&lt;/span&gt;&lt;/sub&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 20 Aug 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/application-development/2026-gartner-mq-for-cloud-native-application-platforms/</guid><category>GKE</category><category>Serverless</category><category>Application Development</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Google is a Leader in the 2026 Gartner® Magic Quadrant™ for Cloud-Native Application Platforms</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/application-development/2026-gartner-mq-for-cloud-native-application-platforms/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Ines Envid</name><title>Sr. Director, Product Management</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Lisa Shen</name><title>Product Manager, Google Cloud</title><department></department><company></company></author></item><item><title>Introducing the Developer Device Platform for agentic mobile app development</title><link>https://cloud.google.com/blog/topics/developers-practitioners/announcing-developer-device-platform-on-google-cloud/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Most enterprises connect with their customers through a device. Whether it’s using a mobile app to order a product, contact customer service, view content, or manage their account, the customer experience depends on how well an app can run locally on the customer’s device.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For this reason, building and testing applications across a wide variety of devices is critical for any enterprise launch that involves locally running components. However, procuring and hosting devices at scale is expensive and complex, and tests are often flaky, inconclusive, or just difficult to debug. This leaves many developers to test launches on the physical phones in their pockets and hope the results apply to most devices. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To solve this challenge, today we are excited to announce the public preview launch of &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Developer Device Platform (DDP) &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;on Google Cloud. DDP is a fully managed cloud platform that provides instant, on-demand access to multiple hardware profiles across real physical devices and high-concurrency virtual emulators. DDP represents an evolution of Firebase Test Lab for Cloud developers, and is also the first device platform built for agentic development. With DDP, developers can now utilize their preferred agents to vibe code apps, run tests, debug, and optimize performance across devices efficiently and quickly.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Build and test your apps to guarantee performance&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In the standard mobile development lifecycle, developers iteratively build new features, run QA tests to ensure performance across a variety of target devices, and ship the optimized and debugged feature to production for their users. Developer Device Platform offers two main functions to accelerate this cycle:&lt;/span&gt;&lt;/p&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Interactive debugging with Device Streaming: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;With our Device Streaming API, developers can directly access an emulator or physical device of their choice, and vibe code, iteratively test, debug and interact with the app remotely. Device streaming makes it simple to dive into your customer experience, and scroll and click in real time, all while also monitoring performance on the real device hardware.&lt;/span&gt;&lt;/p&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Parallel testing with Device Run: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;With our Device Run API, developers can write tests as part of their CI/CD pipelines and run them in parallel across hundreds of different devices at once. With the results, developers can pinpoint and debug specific device issues, and ship code to production with confidence that it will run across device tiers.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Accelerate mobile app development with DDP agent skills and efficient tests&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The rise of coding agents in mobile development specifically opens up new possibilities when paired with physical devices. Coding agents can interact and test on the real hardware, helping them take advantage of unique phone screen sizes (e.g., foldable phone UI) and specialized hardware (e.g., CPUs vs GPUs). &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Developer Device Platform will soon integrate with Android Studio and Android CLI, giving you direct access to physical devices via Device Streaming API. The DDP agent skill will also allow you to work with the AI coding agents of your choice to accelerate development. With &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/developer-device-platform/agent-skill"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;DDP agent skill&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, coding agents can:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Execute multi-step user journeys independently&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Spot visual artifacts&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;Analyze real-time chip performance on-device&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;Validate fixes to hardware specific bugs and/or optimize for unique phone features&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In addition to these agentic capabilities, DDP also enables developers to package apps and launch parallelized tests with smart sharding, giving you access to results across hundreds of devices in minutes. With smart auto-retries, DDP also retries specific tests that fail within your shards, helping you get past errors faster without rerunning your entire suite of tests.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Start building with Developer Device Platform today&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Starting August 12, Developer Device Platform is available in public preview to all Google Cloud users. During public preview, users are charged based on a pay-per-minute model so you pay only for the active testing minutes you consume, with rates differing for emulator vs physical devices.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We can’t wait to see how Developer Device Platform can help mobile developers across Google Cloud accelerate their development and take advantage of the growing number of unique device features and on-device AI possibilities. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Mon, 10 Aug 2026 22:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/developers-practitioners/announcing-developer-device-platform-on-google-cloud/</guid><category>Application Development</category><category>Developers &amp; Practitioners</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_-_Header_image.max-600x600.jpg" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Introducing the Developer Device Platform for agentic mobile app development</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/1_-_Header_image.max-600x600.jpg</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/developers-practitioners/announcing-developer-device-platform-on-google-cloud/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Derek Bekebrede</name><title>Product Manager, Google</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Jason Nager</name><title>Product Strategy &amp; Operations</title><department></department><company></company></author></item><item><title>Frontier and Center: Who evaluates the evaluations?</title><link>https://cloud.google.com/blog/products/data-analytics/evaluate-agent-performance/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;strong style="font-style: italic; vertical-align: baseline;"&gt;Editor’s note:&lt;/strong&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt; Some of the most interesting questions in AI are being asked by information theoreticians, around how to provide context to an emerging class of AI agents. A few weeks ago, we waded into those waters with a blog about &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing?e=0"&gt;&lt;span style="font-style: italic; text-decoration: underline; vertical-align: baseline;"&gt;the Open Knowledge Format&lt;/span&gt;&lt;/a&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;, a specification that formalizes the LLM-wiki pattern into a portable, interoperable format to represent the metadata, context, and curated knowledge that modern AI systems need to operate. That blog generated a ton of interest, so we’ve decided to bring you more of the same, as part of our new “Frontier and Center” series. Today, we hear from two members of Google Data Cloud’s frontier AI team on the recurring challenge of how to systematically evaluate whether or not an agent is able to answer questions effectively based on its context. Read on for more, and watch this space for more blogs from this team.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;A passing grade is the least interesting thing an exam can tell you. It says the student cleared the bar; leaving you entirely in the dark about how narrow their failures were, how effortless their passes were, or what to teach next. Yet this is exactly how we evaluate AI agents. We run a fixed benchmark, calculate a score, and declare progress. In doing so, we are handing our agents a pass/fail exam when what we actually need is a map of the agent’s capabilities: a picture of the terrain that shows exactly where capability falls off, and by how much.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For data agents, this map matters a lot for data discovery&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;in search and retrieval — the unglamorous first step where an agent, handed a vague human question and a warehouse or data lake of thousands of tables and files, has to find the &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;right&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; datasets before it can reason over anything. Discovery is a "needle in a haystack" problem. Real users phrase their questions imperfectly, and inferring what datasets to retrieve presents a real challenge to agents. So the interesting question in evaluations is never "can the agent pass?" It is "how vague can the question get before the agent breaks?" An exam cannot easily answer that, but a map can. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Today, we share an approach rooted in information theory that we’ve been leveraging to add detail and nuance, i.e., fidelity, to benchmarks, so we can better understand agents’ performance as a part of their evaluations. Along the way, the added fidelity exposed some deeper issues with the quality of emergent evaluation cases themselves.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Difficulty, measured&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;When it comes to retrieval, evaluation cases are often stratified into tiers of difficulty. This can happen organically, e.g., pervasive and enduring failure scenarios are deemed difficult. Or it can be from labels applied by humans or machines categorizing some questions as "easy" or "hard" for an agent to answer correctly, e.g., based on the context provided in the query. While this kind of sentiment-based labeling is not the only way to label test cases, it’s frequently used despite its imperfections, such as being challenging to reproduce.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Despite being an industry staple, the approach of assessing every evaluation case by hand is unrealistic at scale. What we need is a rigorous approach that can modulate the difficulty of evaluation cases. We’re iterating on a &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;meta-benchmark&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; we call Discovery Bench: a framework that modulates an evaluation case by generating “easy” and “hard” variations of every case. This allows us to audit how close or how far an agent is from succeeding in those cases. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The lever for modulating the difficulty of an input query comes via a tried-and-trusted concept that’s present across information theory and machine learning: surprisal, or the likelihood of an output given a set of inputs. In our case, a query’s surprisal represents the uncertainty that remains about the correct dataset given the query.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The thinking behind our approach is simple: A term or a phrase in an evaluation query has high informative power when it sharply distinguishes the target from everything else in the corpus. Therefore, we can adjust the difficulty of evaluation cases by adding or removing terms with varying levels of informative power.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Let’s work through a real example from &lt;/span&gt;&lt;a href="https://github.com/mitdbg/KramaBench" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;KramaBench&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a publicly available benchmark. One of KramaBench’s datasets has information about orbiting satellites, and the example query from the suite includes the following text: &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;"…the total count of satellite major altitude changes for satellite 48445 during 2024 using TLE history."&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The token &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;"TLE"&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; is sharply distinguishing; it points almost uniquely at the &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;TLE_____48445&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; table from the dataset. Strip it, and the query degrades to &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;"the count of satellite altitudes for satellite 48445,"&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; whose vague phrasing now matches density tables, precise-orbit files, and decay logs alike. Surprisal makes this quantitative: rare, pointed terms carry more bits than common ones.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The remaining surprisal of a query is how much uncertainty is left about its answer. As surprisal approaches zero&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;,&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; the query has become specific enough to pinpoint exactly one dataset.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The heart of the idea behind Discovery Bench is this refinement loop, which we call &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;iterative surprisal-based query refinement&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;, or iSQR, which generates cases with higher or lower informative power to test where an agent can start successfully answering the query:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/Who_Evaluates_the_Evaluations__-_FP_blog.max-1000x1000.jpg"
        
          alt="Who Evaluates the Evaluations_ - FP blog"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="noe01"&gt;Figure 2: The iSQR refinement loop.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The crux is being able to control the challenge embedded into the evaluation case by making adjustments: Instead of one fixed phrasing per question, we generate the &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;same&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; question at three levels of calibrated ambiguity [high, medium, low], with each grounded in bits (not subjective opinion). We can even justify, term by term, &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;why&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; a word was added or removed. Difficulty stops being a property that is attributed by sentiment or classification, and becomes one we &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;engineer&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The cliff you couldn't see&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Here is what Discovery Bench’s difficulty dial reveals — and what a single-phrasing benchmark structurally cannot.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We have an F1 agent that's built for recall (on Gemini 3.1 Pro). Running it against KramaBench and across the full sweep of ambiguity levels traces a curve: 0.34 at high ambiguity, 0.76 at neutral, 0.81 at medium, 0.78 at low.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/image1_viEBS6x.max-1000x1000.jpg"
        
          alt="image1"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="noe01"&gt;Figure 3: F1 swept across ambiguity — the dot versus the curve.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Two findings fall out immediately (and neither were visible to a conventional eval).&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;First, the cliffs.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; This query scores a perfect &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;F1 = 1.00&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; at neutral phrasing — and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;0.00&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; at high ambiguity. It is the satellite-48445 case from above: drop the distinguishing token "TLE" and the agent loses the table entirely. Same query, same agent, same ground truth; one notch vaguer and it falls off a cliff. A static benchmark tests the neutral phrasing, stamps "solved," and reports flat ground where there is a precipice. Pass/fail was particularly misleading in that it did not just miss the cliff, but it told us the terrain was level.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Second, the sweet spot.&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; For Discovery Agent, medium ambiguity &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;beat&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; neutral, and low ambiguity sometimes &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;underperformed&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; it. More specificity is not monotonically better for the system being evaluated; there is an optimal amount of steering. That is a graded, actionable signal. This is the "how close, how hard" texture we were missing from a scalar. It tells you where to hill-climb, or improve, the agent: in our case, straight at concrete failure modes like time-sharded tables (precision collapsing to ~8% as the agent over-retrieves 21 near-identical shards for a two-table answer) and context blow-up (F1 dropping from 0.75 to 0.32 once a query triggers long search chains). The map did not just say that the agent failed, but it said where, and why. Note that our hypothesis that less ambiguity and more context (via steering terms) should improve retrieval generally holds true, but for the specific Discovery Agent being exercised, the idiosyncratic “sweet spot” meaningfully highlighted trade-offs in its implementation.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;We're not alone&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The field is converging on meta-benchmarking and exerting greater control of how we challenge and evaluate our agents. A growing body of work uses &lt;/span&gt;&lt;a href="https://en.wikipedia.org/wiki/Item_response_theory" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;item response theory&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, the latent-ability model behind standardized testing, to treat difficulty as a measured quantity rather than a label: &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2402.14992" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;tinyBenchmarks&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2407.12844" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;metabench&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; show that a handful of informative items reproduce a model's full score, and &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2505.15055" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;PSN-IRT&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; turns the same lens on benchmark quality itself. Others audit the ground truth directly: &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2406.04127" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;MMLU-Redux&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; found that 6.49% of Massive Multitask Language Understanding (MMLU) questions are mislabeled, and &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2502.03461" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Platinum Benchmarks&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; re-cleaned ten datasets to minimize both label errors and ambiguity — the same two axes we sweep for. And ambiguity is increasingly treated as intrinsic rather than noise: &lt;/span&gt;&lt;a href="https://aclanthology.org/2020.emnlp-main.466/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;AmbigQA&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; showed that a large fraction of real questions admit multiple readings, and later work finds that apparent hallucinations often stem from query ambiguity rather than model failure. What we have not seen elsewhere is the combination: information-theoretic ambiguity sweeping applied as a meta-benchmark over live enterprise data.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;A benchmark we trusted turned out to be broken&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We built our first evaluation on &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2506.06541" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;kramabench-astronomy&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a benchmark established in the field, and one which other teams had already leaned on for their own evals. Teams derived benchmarks from this dataset, and we hypothesized subtle issues may have been introduced over time. When we actually read the benchmarks used by teams, with Gemini's help, we found it was wrong in meaningful ways: ground-truth tables that did not answer their query, a question whose 124 sharded tables exceeded what some teams’ retrieval APIs could even return, months specified where exact dates were required. Quietly broken ground truth means quietly wrong conclusions not just for us, but for every prior analysis built on it. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This is the generalized crux of the matter: an evaluation is itself an artifact that can be defective, and almost nobody evaluates it. We instrument the agent and trust the ruler, but where do we validate that the measuring stick makes sense? &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;When two maps disagree&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Now the recursive turn: If difficulty is something we &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;generate&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;, then we need to evaluate the generator itself; we should not trust it blindly either.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;So we built the &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;same&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; ambiguity sweep two ways: steering terms from a pure-LLM guess, versus terms grounded in &lt;/span&gt;&lt;a href="https://en.wikipedia.org/wiki/Tf%E2%80%93idf" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;TF-IDF surprisal&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. The two disagreed violently. At high ambiguity, the LLM-built sweep scored the agent at F1 ≈ &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;0.34&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;; the grounded sweep, ≈ &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;0.85&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. One of these maps is badly distorted. The grounded one, predictably, is the more robust: surprisal gives it a footing the free-running LLM lacks.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This is "evaluate your evals," made concrete. The information-theoretic lens does not only grade the agent along a continuous axis; it grades the &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;benchmark's own construction&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;, and adjudicates between the two.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Evaluate your evals&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We have spent years optimizing agents against rulers we never measured. The bitter irony is that better models make this worse: as agents clear coarse benchmarks, the score saturates near the top and the exam loses its ability to highlight where the agent can be improved.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;So the call to action is uncomfortable and overdue: evaluate your evals. Read your ground truth. Treat difficulty as a measured quantity, not a label: sweep it, plot it, find the bit-width where your system breaks. Ask not just "did it pass?" but "how close was the miss, how hard was the pass, and would a slightly vaguer question have sent it off a cliff?" Build evaluations that produce signals; not just verdicts.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;There is a genuine tension to sit with here. Difficulty-as-entropy is only as reliable as the model that estimates the entropy. There's a risk that if we push too hard on a measurable proxy, we optimize the ruler instead of the agent. That is not a reason to retreat to pass/fail; it is a reason to keep the evaluator under the same scrutiny as what it is evaluating. The moment we stop asking who evaluates the evaluators is the moment our maps stop being useful again.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p role="presentation"&gt;&lt;sup&gt;&lt;em&gt;&lt;span style="vertical-align: baseline;"&gt;1. Maia Polo, F. et al. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;tinyBenchmarks: Evaluating LLMs with Fewer Examples.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; ICML 2024. &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2402.14992" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;arxiv.org/abs/2402.14992&lt;/span&gt;&lt;/a&gt;&lt;br/&gt;&lt;/em&gt;&lt;/sup&gt;&lt;sup&gt;&lt;em&gt;&lt;span style="vertical-align: baseline;"&gt;2. Kipnis, A. et al. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;metabench: A Sparse Benchmark of Reasoning and Knowledge in Large Language Models.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; ICLR 2025. &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2407.12844" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;arxiv.org/abs/2407.12844&lt;/span&gt;&lt;/a&gt;&lt;br/&gt;&lt;/em&gt;&lt;/sup&gt;&lt;sup&gt;&lt;em&gt;&lt;span style="vertical-align: baseline;"&gt;3. Lost in Benchmarks? Rethinking Large Language Model Benchmarking with Item Response Theory&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; (PSN-IRT). AAAI 2026. &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2505.15055" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;arxiv.org/abs/2505.15055&lt;/span&gt;&lt;/a&gt;&lt;br/&gt;&lt;/em&gt;&lt;/sup&gt;&lt;sup&gt;&lt;em&gt;&lt;span style="vertical-align: baseline;"&gt;4. Gema, A. P. et al. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Are We Done with MMLU?&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; (MMLU-Redux). 2024. &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2406.04127" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;arxiv.org/abs/2406.04127&lt;/span&gt;&lt;/a&gt;&lt;br/&gt;&lt;/em&gt;&lt;/sup&gt;&lt;sup&gt;&lt;em&gt;&lt;span style="vertical-align: baseline;"&gt;5. Vendrow, J. et al. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Do Large Language Model Benchmarks Test Reliability?&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; (Platinum Benchmarks). 2025. &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2502.03461" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;arxiv.org/abs/2502.03461&lt;/span&gt;&lt;/a&gt;&lt;br/&gt;&lt;/em&gt;&lt;/sup&gt;&lt;sup&gt;&lt;em&gt;&lt;span style="vertical-align: baseline;"&gt;6. White, C., Dooley, S. et al. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;LiveBench: A Challenging, Contamination-Limited LLM Benchmark.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; 2024. &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2406.19314" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;arxiv.org/abs/2406.19314&lt;/span&gt;&lt;/a&gt;&lt;br/&gt;&lt;/em&gt;&lt;/sup&gt;&lt;sup&gt;&lt;em&gt;&lt;span style="vertical-align: baseline;"&gt;7. Min, S. et al. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;AmbigQA: Answering Ambiguous Open-domain Questions.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; EMNLP 2020. &lt;/span&gt;&lt;a href="https://aclanthology.org/2020.emnlp-main.466/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;aclanthology.org/2020.emnlp-main.466&lt;/span&gt;&lt;/a&gt;&lt;br/&gt;&lt;/em&gt;&lt;/sup&gt;&lt;sup&gt;&lt;em&gt;&lt;span style="vertical-align: baseline;"&gt;8. Lai, E., Vitagliano, G. et al. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;KramaBench: A Benchmark for AI Systems on Data-to-Insight Pipelines over Data Lakes.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; 2025. &lt;/span&gt;&lt;a href="https://arxiv.org/abs/2506.06541" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;arxiv.org/abs/2506.06541&lt;/span&gt;&lt;/a&gt;&lt;/em&gt;&lt;/sup&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Fri, 10 Jul 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/data-analytics/evaluate-agent-performance/</guid><category>AI &amp; Machine Learning</category><category>Application Development</category><category>Data Analytics</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Frontier and Center: Who evaluates the evaluations?</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/data-analytics/evaluate-agent-performance/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Manav Garg</name><title>Software Engineer, Data Cloud Frontier AI</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Sunil Pedapudi</name><title>Technical Lead, Data Cloud Frontier AI</title><department></department><company></company></author></item><item><title>Boost BigQuery with Python: Managed Python UDFs now generally available</title><link>https://cloud.google.com/blog/products/data-analytics/python-udf-in-bigquery-now-generally-available/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;SQL is the industry standard for high-performance structured data analysis. However, expressing complex procedural logic, scientific computations, advanced string manipulations, or machine learning workflows in pure SQL can be highly challenging, if not impossible. That kind of work is better done with Python. Data practitioners often take on additional infrastructure management tasks &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;—&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; maintaining custom images and containers, and working with additional compute services — just to run simple helper functions with custom Python code and libraries. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span&gt;&lt;span style="vertical-align: baseline;"&gt;Today, we are thrilled to announce the general availability (GA) of&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;a href="https://docs.cloud.google.com/bigquery/docs/user-defined-functions-python"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;BigQuery Managed Python User-Defined Functions (UDFs)&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This launch represents a major milestone in BigQuery’s extensibility strategy, allowing data scientists, engineers, and analysts to execute custom Python code directly and securely inside BigQuery using standard SQL queries or &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/bigquery/docs/bigquery-dataframes-introduction"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;BigQuery DataFrames&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (BigFrames) in Python. With this release, Python UDFs are fully supported for production enterprise workloads and completely integrated into BigQuery's billing SKUs. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Bridging SQL and the Rich Python Ecosystem&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;BigQuery Managed Python UDFs run on BigQuery-managed serverless resources that automatically scales to billions of rows, without having to set up infrastructure or manage containers. BigQuery automatically handles the compilation, image building, security patching, deployment, and execution of your Python code, making it super simple to use Python functions in your SQL.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Core benefits&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Flexibility:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Access the vast Python ecosystem — including top-tier scientific and mathematical libraries like NumPy, SciPy, pandas, scikit-learn and more — directly in your SQL select statements.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Tight external API integration:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Clean and enrich your BigQuery tables in real time by calling external web APIs or Google Cloud services such as Cloud Translation, Gemini Enterprise Agent Platform or custom microservices securely within your queries.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Fully managed and serverless:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; BigQuery handles the underlying container infrastructure and auto-scales performance dynamically.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Code example &lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Here is an example of a Python UDF that utilizes a popular Python package —&lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt; beautifulsoup&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; — to remove HTML tags. We use this function to process &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;StackOverflow answer bodies that are stored in a BigQuery public table:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;CREATE OR REPLACE FUNCTION `your_project.your_dataset.clean_html`(html_content STRING)\r\nRETURNS STRING\r\nLANGUAGE python\r\nOPTIONS (\r\n  runtime_version = \&amp;#x27;python-3.11\&amp;#x27;,\r\n  entry_point = \&amp;#x27;strip_tags\&amp;#x27;,\r\n  packages = [\&amp;#x27;beautifulsoup4&amp;gt;=4.12.0\&amp;#x27;]\r\n) AS r\&amp;#x27;\&amp;#x27;\&amp;#x27;\r\nfrom bs4 import BeautifulSoup\r\n\r\ndef strip_tags(html_content):\r\n    if not html_content:\r\n        return &amp;quot;&amp;quot;\r\n    soup = BeautifulSoup(html_content, &amp;quot;html.parser&amp;quot;)\r\n    return soup.get_text(separator=&amp;quot; &amp;quot;)\r\n\&amp;#x27;\&amp;#x27;\&amp;#x27;;&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4432e90&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;How to query it:&lt;/strong&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;SELECT \r\n  id, \r\n  `your_project.your_dataset.clean_html`(body) AS cleaned_answer_body\r\nFROM \r\n  `bigquery-public-data.stackoverflow.posts_answers`\r\nLIMIT 100&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5ccf826010&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Advanced capabilities&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For advanced users, Python UDF adds a set of capabilities to tune the performance as well as monitor the usage. Here are some examples. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Vectorized processing with Pandas PyArrow&lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;To maximize throughput, the GA release supports direct processing of vectorized input as PyArrow RecordBatches. By processing columns of data in bulk rather than row-by-row, PyArrow eliminates Python serialization and conversion overhead, boosting performance by up to 10x for data-intensive calculations.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Configurable container resources&lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;For heavy-duty data science and ML data preparation, you can now provision container memory (up to 16 GB) and CPU (up to 4 vCPUs) per function. This enables memory-intensive workloads (such as loading large serialized models or geospatial datasets) to run directly within the sandbox.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Customizable concurrency&lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Optimize your throughput and resource efficiency by configuring concurrent requests per container (up to 1,000 concurrent operations). This helps ensure that your scale-out execution is highly cost-effective and performs exceptionally well under heavy parallel loads.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Streaming logs and real-time metrics&lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Easily d&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;ebug and monitor your production workloads. The BigQuery console now features a direct link from your query results to real-time CPU, memory, and concurrency metrics in Cloud Monitoring.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Billing&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;BigQuery Managed Python UDF are billed with &lt;/span&gt;&lt;a href="https://cloud.google.com/bigquery/pricing#bigquery-services-pricing"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;BigQuery Services SKU&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. This SKU is fully eligible for &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;BigQuery spend commitment-based usage discounts (CUDs)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, allowing you to maximize budget efficiency.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;You can also get cost observability through &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;INFORMATION_SCHEMA.JOBS &lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;as well as using billing labels &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;MANAGED_ROUTINE_EXECUTION&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;MANAGED_ROUTINE_BUILD&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;).&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;See more details in the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/bigquery/docs/user-defined-functions-python#pricing"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Pricing&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; section of the documentation. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Getting started &lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To get started with BigQuery Python UDFs, first check out &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/bigquery/docs/user-defined-functions-python"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;product documentation&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Then, try out the functions &lt;/span&gt;&lt;a href="https://console.cloud.google.com/bigquery?ws=!1m5!1m4!6m3!1sbigquery-public-data!2spython_udfs!3stokenize"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;published&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in the public BigQuery dataset. For example, run the following code in a BigQuery project to tokenize country names data from BigQuery public data. Under the hood, the token UDF utilizes the &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;o200k_base&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; tokenizer library.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;SELECT \r\n  country_code,\r\n  country_name,\r\n  `bigquery-public-data`.python_udfs.tokenize(country_name) AS name_tokens,\r\n  ARRAY_LENGTH(`bigquery-public-data`.python_udfs.tokenize(country_name)) AS token_count\r\nFROM \r\n  `bigquery-public-data.census_bureau_international.country_names_area`\r\nORDER BY \r\n  country_name&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd511dc10&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Or, try out this &lt;/span&gt;&lt;a href="https://codelabs.developers.google.com/managed-python-udfs" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;code lab&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to explore some advanced scenarios. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Then, to learn how to implement other advanced design patterns, we encourage you to explore our official public documentation guides: &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Calling Google Cloud or online services (with connections):&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; To connect to first-party Google Cloud services such as Gemini Enterprise Agent Platform or Cloud Translation, or external API endpoints securely using Cloud Resource connections, - check out the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/bigquery/docs/user-defined-functions-python#use-online-service"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Call Google Cloud or online services in Python code guide&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;BigQuery DataFrames (BigFrames) Python UDFs:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;To learn how to write, deploy, and scale custom Python functions natively from standard Jupyter notebook or Colab environments using BigQuery DataFrames, visit the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/bigquery/docs/user-defined-functions-python#bigquery-dataframes_1"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Customize Python functions for BigQuery DataFrames guide&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Bring your Python workflows out of isolation and directly into the heart of your data warehouse today!&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Mon, 22 Jun 2026 17:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/data-analytics/python-udf-in-bigquery-now-generally-available/</guid><category>Application Development</category><category>Data Analytics</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Boost BigQuery with Python: Managed Python UDFs now generally available</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/data-analytics/python-udf-in-bigquery-now-generally-available/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Sandeep Karmarkar</name><title>Group Product Manager</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Chao Shen</name><title>Tech lead</title><department></department><company></company></author></item><item><title>Go from resource-level to business-level maintenance in Google Cloud</title><link>https://cloud.google.com/blog/products/application-development/app-centric-maintenance-visibility-in-unified-maintenance/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Managing planned maintenance is a critical part of running a reliable business. But as your cloud footprint grows into hundreds or even thousands of projects, keeping track of every individual update can feel like a full-time job. For many platform teams, the reality is a disjointed experience: jumping between dashboards to figure out which maintenance event affects which business service.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;At Google Cloud, we believe you shouldn’t have to think like an infrastructure manager when you’re trying to solve a business problem. That’s why we are excited to announce the launch of &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;App-centric maintenance visibility&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; within &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/unified-maintenance/docs/overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Unified Maintenance&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Shifting the focus to your business&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Until now, maintenance visibility was primarily resource-focused. You could see when a specific Compute Engine VM or Cloud SQL instance was due for an update, but you had to manually map those resources to the applications they powered.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With App-centric visibility, we are shifting the focus from infrastructure-level resources to a business-oriented view. By integrating directly with &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;App Hub&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, Unified Maintenance now allows you to see maintenance events in the context of your applications.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;How it works&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This new capability leverages the "application" as the primary unit of management. When you register your resources in App Hub — whether they are GKE clusters, GCE VMs, or AlloyDB instances—Unified Maintenance automatically aggregates their maintenance schedules into a single, application-aware dashboard.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_eoV1hGN.max-1000x1000.png"
        
          alt="1"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/2_O2U4S4N.max-1000x1000.png"
        
          alt="2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For platform engineers, this means:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Reduced toil&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: No more manual mapping of maintenance alerts to application owners.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Faster triage&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Instantly see if a performance dip in an app coincides with a planned infrastructure update&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Predictable operations&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Gain a business-oriented view of maintenance impacts across your entire landscape &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Get started today&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span&gt;&lt;span style="vertical-align: baseline;"&gt;Enable the &lt;/span&gt;&lt;a href="http://console.cloud.google.com/apis/library/maintenance.googleapis.com"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Maintenance API&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; now. If you already have &lt;/span&gt;&lt;a href="https://console.cloud.google.com/apphub/applications"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Applications&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; defined in Google cloud, you can explore the new App-centric visibility features directly in the &lt;/span&gt;&lt;a href="https://console.cloud.google.com"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud Console&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. To learn more about setting up your application boundaries and mapping resources, check out our &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/unified-maintenance/docs/set-up-unified-maintenance"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Get Started Guide&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Finally, don’t forget to check out our &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/unified-maintenance/docs/supported-services"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;supported services page&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; as we are onboarding new services.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 28 May 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/application-development/app-centric-maintenance-visibility-in-unified-maintenance/</guid><category>Application Development</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Go from resource-level to business-level maintenance in Google Cloud</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/application-development/app-centric-maintenance-visibility-in-unified-maintenance/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Erol-Valeriu Chioasca</name><title>Product Manager, Google Cloud</title><department></department><company></company></author></item><item><title>Shipping features to production just got easier with new feature flags in AppLifecycle Manager</title><link>https://cloud.google.com/blog/products/application-development/new-feature-flags-in-applifecycle-manager/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Many development teams are familiar with the hesitation that comes right before pushing a new feature live. As AI helps developers write code faster, the gap between rapid code generation and safe production deployment continues to grow.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Feature flags offer a practical way to manage this risk by separating the act of deploying code from the act of releasing a feature to users. Instead of a single, high-risk launch event that affects all users simultaneously, teams can ship code to production with new features hidden by default in a controlled manner.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To help teams adopt this workflow, we are announcing the public preview of AppLifecycle Manager Feature Flags (ALM FF). This service provides a rule-based solution to manage software behavior across Google Cloud, helping you support rapid development without sacrificing production stability.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Read on to learn four ways these feature flags will help accelerate your deployment.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;1. Decouple for safety and velocity&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The core mission of ALM FF is to increase development velocity by decoupling your feature releases from your code deployments. Traditionally, releasing a feature requires a binary deployment — a high-risk event that affected all users simultaneously.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With ALM FF, you can ship code to production with new features disabled by default. This allows your team to move faster, deploying code continuously while choosing the exact moment to enable a feature via a toggle. If an issue is detected, the flag acts as an instant kill switch, disabling the problematic feature immediately without the need for a full, time-consuming code rollback.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_YPP5fhI.max-1000x1000.png"
        
          alt="1"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;2. Gradual enablement with precise targeting&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Safety is  about precision. ALM FF leverages the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Common Expression Language (CEL)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; to implement sophisticated logic for gradual feature enablement.&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Percentage feature enablement:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Instead of a global launch, you can ramp up a feature to 1%, 5%, or 50% of your traffic. This allows you to monitor system health and performance metrics incrementally, ensuring stability before reaching your entire user base.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Precise allowlisting:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; You can target specific internal teams, beta testers, or early-access customers by allowlisting their identifiers. This ensures that only the intended audience sees a feature during its initial validation phase.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/2_AT9I4Zt.max-1000x1000.png"
        
          alt="2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;3. Dynamic configuration for the AI era&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Beyond simple toggles, ALM FF offers a dynamic way to inject configuration into your applications. By using string-type flags, you can update application behavior — such as system prompts for LLM integrations—in real-time. This allows product managers and business owners to tweak AI responses and application logic without requiring any code changes or infrastructure rollouts.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/3_7li94CT.max-1000x1000.png"
        
          alt="3"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;4. Built on open standards&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We believe safety should not mean lock-in. ALM FF is built on the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;OpenFeature&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; standard, utilizing industry-standard SDKs and the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;flagd&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; evaluation engine. This ensures your feature management patterns are portable and follow best practices without adding Google-specific dependencies to your core application code.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Get started&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;ALM FF is now in public preview. To take control of your releases, you can:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Review the docs:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/saas-runtime/docs/flags/flags-overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Public Documentation&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Onboard today:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/saas-runtime/docs/flags/flags-quickstart"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Quickstart Guide&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Give us feedback:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Help us &lt;/span&gt;&lt;a href="https://forms.gle/boGXCgKyoB7Lr6yd9" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;shape the future of feature management&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;</description><pubDate>Thu, 21 May 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/application-development/new-feature-flags-in-applifecycle-manager/</guid><category>Developers &amp; Practitioners</category><category>Application Development</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Shipping features to production just got easier with new feature flags in AppLifecycle Manager</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/application-development/new-feature-flags-in-applifecycle-manager/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Erol-Valeriu Chioasca</name><title>Product Manager, Google Cloud</title><department></department><company></company></author></item><item><title>AI Studio unlocks full-stack vibe coding with Cloud Run, Firebase, and Cloud SQL, no credit card required</title><link>https://cloud.google.com/blog/products/databases/vibe-coded-ai-studio-apps-with-firestore-firebase-cloud-sql/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;At&lt;/span&gt;&lt;a href="https://io.google/2026/" rel="noopener" target="_blank"&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google I/O 2026&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, we announced  updates to the integration between &lt;/span&gt;&lt;a href="https://aistudio.google.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google AI Studio&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://cloud.google.com/"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;: &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;New users can deploy up to two full-stack applications to the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/docs/starter-tier"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud Starter Tier, &lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;no billing account required&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;An expanded choice of databases: &lt;/span&gt;&lt;a href="https://cloud.google.com/products/firestore"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Firestore&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; for non-relational data, and &lt;/span&gt;&lt;a href="https://cloud.google.com/sql/postgresql"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud SQL&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; as a new relational database option&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Tight integration with Google Workspace tools like Sheets, Calendar, and Gmail using &lt;/span&gt;&lt;a href="https://firebase.blog/posts/2026/05/google-io-2026-announcements" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Firebase Auth as the single user login flow&lt;/span&gt;&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This is an update to &lt;/span&gt;&lt;a href="https://blog.google/innovation-and-ai/technology/developers-tools/full-stack-vibe-coding-google-ai-studio/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;the integration we announced&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in March, which included support for vibe-coded full-stack app deployments from AI Studio powered by &lt;/span&gt;&lt;a href="https://cloud.google.com/run"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Run&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;a href="https://firebase.blog/posts/2026/03/announcing-ai-studio-integration" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Firestore, and Firebase Auth&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With this expanded integration, you can use AI Studio to build a broader set of applications, using either a relational database with Cloud SQL or a non-relational database with Firestore. You don’t even need to specify a database — the AI agent can infer the right database for your app or feature.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href="https://aistudio.google.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Get started today&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in AI Studio at no cost with Cloud Run, Cloud SQL for PostgreSQL (coming next month), Firestore, and Firebase Auth for Starter Tier.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/1-_publish.gif"
        
          alt="1- publish"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="3iru6"&gt;Publishing a full-stack app from AI Studio to Cloud Run with a single click&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;An easy on-ramp: The Google Cloud Starter Tier&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;You can build applications in AI Studio and deploy your prototypes directly to Cloud Run, authenticate via Firebase Auth, and store your data in a Firestore or Cloud SQL database. No credit card, no Google Cloud account, no friction — just prompt and launch.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;If you don’t have an account, AI Studio uses the Google Cloud Starter Tier to create resources for you. You can deploy up to two full-stack apps. If you outgrow the limits of the Starter Tier, you can upgrade to a standard Google Cloud project with a billing account. All your resources will be transferred to your billable Google Cloud project, so that your application can scale as it grows.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Powering full-stack vibe coding with Cloud SQL&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We’re introducing an intelligent, automated data foundation that makes it easy for developers to focus on their applications, not their infrastructure.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;AI Studio integration with Cloud SQL includes:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;An instant on-ramp:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Go from prompt to a fully-deployed PostgreSQL database rapidly with instant provisioning.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Zero-cost startup:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Try Cloud SQL for the Google Cloud Starter Tier at no cost, without needing a credit card or Google Cloud account. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Flexible cost control:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; The AI agent uses a new Cloud SQL for PostgreSQL developer edition, which enables the backend to scale to zero automatically, so you only pay while you’re using the app.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent-driven experience:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; To update your application, enter new prompts and the AI Agent automatically creates the schema and executes SQL statements in the database.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Global scalability:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; While the interface is simple, your app runs on Google Cloud’s robust, highly-reliable, and securely designed infrastructure that can scale to support millions of users.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/4_-_Cloud_SQL_AIS_Demo.gif"
        
          alt="4 - Cloud SQL AIS Demo"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="3iru6"&gt;Creating an app powered by Cloud SQL for PostgreSQL developer edition&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Full-stack vibe coding with Firestore and Firebase Auth&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;When you’re building an app in AI Studio, the agent proactively detects if you need data storage and authentication based on your prompt, and offers to set up a database and user authentication. For apps that benefit from a document database, the agent shows a card to turn on Firestore and Firebase Authentication with your approval. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/image3_BDw1RGs.max-1000x1000.png"
        
          alt="2-enable firebase"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="3iru6"&gt;Enable Firebase for your application when prompted by the agent&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;By clicking “Enable Firebase,” the agent automatically:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Provisions Firestore, enables authentication, and connects your app to the database&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Creates your web app’s sign-in page and configures authentication with Google Sign In&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Generates the Firestore code in your app so you can sync data across sessions and devices&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Drafts and deploys Firestore Security Rules based on your app’s logic (but you should always double-check these rules before sharing or deploying your app!)&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;With Firebase Auth, you can:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Connect your apps to Google Workspace using natural language: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;When you ask for a feature involving Workspace (e.g. Sheets, Calendar, Gmail), the agent implements a “Sign in with Google” flow, powered by Firebase Authentication, designed to securely grant Google AI Studio access to your data.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/image1_MMqzmOz.max-1000x1000.png"
        
          alt="image1"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="jf96o"&gt;Connect your app to Google Sheets, powered by Firebase Authentication&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Check out more details on the &lt;/span&gt;&lt;a href="https://firebase.blog/posts/2026/05/google-io-2026-announcements" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;What’s New from Firebase at Google I/O blog&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Getting started in AI Studio&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Going from&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; idea to app is now a reality. You can build a full-stack application at no cost using the following steps:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Log into AI Studio:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Access the platform to begin your project.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Build with prompts:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Start building your application using natural language prompts. For example, “Build an expense tracker app.”&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Enable the database:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Prompt “Add a database” and AI Studio intelligently provisions a database through an "Enable" widget. You can explicitly ask for a relational database if you’d like to make your preference clear.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Set up the system:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Select “Enable” and agree to the terms.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Start sharing:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Deploy and share the application through the “Publish” button.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Get started today&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; in &lt;/span&gt;&lt;a href="https://aistudio.google.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;AI Studio&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to turn your ideas into live applications in seconds.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 21 May 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/databases/vibe-coded-ai-studio-apps-with-firestore-firebase-cloud-sql/</guid><category>Application Development</category><category>AI &amp; Machine Learning</category><category>Firebase</category><category>Serverless</category><category>Databases</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>AI Studio unlocks full-stack vibe coding with Cloud Run, Firebase, and Cloud SQL, no credit card required</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/databases/vibe-coded-ai-studio-apps-with-firestore-firebase-cloud-sql/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Justin Mahood</name><title>Product Management</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Gopal Ashok</name><title>Group Product Manager, Databases</title><department></department><company></company></author></item><item><title>Introducing Agent Executor, Google’s distributed Agent Runtime</title><link>https://cloud.google.com/blog/products/ai-machine-learning/agent-executor-googles-distributed-agent-runtime/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As models and harnesses improve, agents are taking on increasingly complex tasks that can run for hours or even days. But as we push agents to do more, this has surfaced a new operational problem: long-running agent workflows are fragile and incredibly hard to manage reliably and efficiently in production.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Today, we’re introducing &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Executor,&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Google’s &lt;/span&gt;&lt;a href="https://github.com/google/ax" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;open-source&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; runtime standard for agent execution, resumption, and distributed deployment. Based on what we’ve learned from solving these challenges internally, we’ve built Agent Executor to have the following native capabilities: &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Durable execution:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Long-running execution requires the ability to resume after outages or agentic interruptions such as human-in-the-loop (HITL) confirmations. Agent Executor provides this backend resilience automatically for any actor (e.g., an agent, agent harness, skill, tool, or sandbox) through its event log and snapshotting.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Secure isolation&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Agent Executor isolates components in secure-by-design sandboxes to prevent harmful side effects and help ensure malicious activity cannot compromise the broader service. Sandboxes are especially useful when agents generate code or handle multiple tenants or user data concurrently.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Session consistency: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;In distributed agent workflows, multiple components may attempt to update shared session state at the same time. Agent Executor’s built-in single-writer architecture helps maintain consistency and reduces the risk of corruption in that state.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Connection recovery:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; In long-running agentic execution, clients may disconnect for many reasons, including network outages. Agent Executor lets clients reconnect to agents and backfills responses from the last sequence seen by the client for a better user experience.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Trajectory branching: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Checkpoints let you branch an agentic trajectory (its decision or workflow path) at any point, allowing agents to test or evaluate different paths without losing context or other state.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_agent_executor_EdQVRpG.max-1000x1000.jpg"
        
          alt="1 agent executor"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In this blog, we’ll share more about Agent Executor and how you can get started. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Federate with Google’s agent runtime&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Enterprise adoption of agents requires orchestration across deployment models. Some teams need on-prem infrastructure for proprietary workflows, performance, or compliance, while others prefer pre-built or custom managed agents for faster time-to-value. At Google I/O, we introduced a new suite of such solutions – including &lt;/span&gt;&lt;a href="https://antigravity.google/blog/introducing-google-antigravity-2-0" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Antigravity 2.0&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and the &lt;/span&gt;&lt;a href="https://blog.google/innovation-and-ai/technology/developers-tools/managed-agents-gemini-api" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Managed Agents API&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; – designed to accelerate how teams build and scale within the agentic enterprise.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Agent Executor bridges these deployment models, letting you mix-and-match between any or all of:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Google &lt;/span&gt;&lt;a href="https://antigravity.google/blog/google-io-2026" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Antigravity&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, Gemini’s state-of-the-art agent harness&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Google-built frontier agents, such as the latest &lt;/span&gt;&lt;a href="https://blog.google/innovation-and-ai/models-and-research/gemini-models/next-generation-gemini-deep-research/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Deep Research&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; agent&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Custom agents built by you and managed by Google (e.g., via the new &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/build/managed-agents"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Managed Agents in Gemini API&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;)&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li&gt;&lt;span style="vertical-align: baseline;"&gt;Custom purpose-built agents, built with LangChain/LangGraph, &lt;/span&gt;&lt;a href="https://adk.dev/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Development Kit&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (ADK), etc and any agents using &lt;/span&gt;&lt;a href="https://github.com/a2aproject/A2A" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent2Agent Protocol&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (A2A)&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/2_sDivt9A.max-1000x1000.jpg"
        
          alt="2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Own your agents, models, and compute&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With Agent Executor, enterprises have maximum flexibility to maintain sovereignty over workloads and keep proprietary workflows within their self-managed compute and custom sandboxes. Your internal development teams have much more flexibility over how agents are deployed and managed and you benefit from:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Prevent vendor lock-in:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Deploy your agents on your own infrastructure without being tethered to a specific provider’s model or compute environment. This allows for full control over data residency and your cost and budgetary controls.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Bring your own harness and agents:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Agent Executor is designed to be harness-agnostic, allowing you to bring your own or use those made available by other vendors. It also supports agents developed with industry-standard frameworks and protocols providing a broad ecosystem of compatible agents.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Fully control execution: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Agent Executor allows developers to run the entire agentic stack, including MCPs, skills, and other agents, directly on their own data plane. Developers can choose any compute with custom isolation boundaries and workload policy enforcement.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Scale agents up on Kubernetes with an agent-first compute layer&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As agent workloads scale into the hundreds of millions and become increasingly long-running, our customers are hitting the limits of traditional compute abstractions because unlike traditional software, agents are nonlinear programs that wait for external inputs. To solve this problem, we’ve partnered with the Google Kubernetes Engine team on &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/containers-kubernetes/bringing-you-agent-sandbox-on-gke-and-agent-substrate"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Substrate&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a new open-source project also announced today.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Agent Substrate introduces a new level of abstraction for Kubernetes that moves agents onto and off of ready compute capacity in real-time, resulting in lower latency with higher scale and efficiency. While standard Kubernetes is optimized to handle thousands of long-running services, Agent Substrate is designed for the chatter of millions of sub-second tool calls that would otherwise overwhelm a standard control plane. Agent Substrate takes core secure runtime and snapshotting capabilities of existing sandbox infrastructure and pairs them with a minimal control plane designed to bypass some of the limitations of Kubernetes, without reinventing the rest of it. Working together, these layers enable you to:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Maximize compute efficiency&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Agent Substrate introduces a new control plane designed to handle hundreds of millions of registered agents. Together with Agent Executor, Agent Substrate can provide a foundation for today’s largest agent deployments.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Stay within the Kubernetes ecosystem:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Agent Substrate is built on top of Kubernetes and allows scheduling and horizontal scaling of compute with declarative configuration.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In the demo below, we showcase using Agent Executor together with Agent Substrate with a sample workload.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=L5Iw1IrZ6Nc"
      data-glue-modal-trigger="uni-modal-L5Iw1IrZ6Nc-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        

        &lt;div class="article-video__aspect-image"
          style="background-image: url(https://storage.googleapis.com/gweb-cloudblog-publish/images/maxresdefault_00CLfta.max-1000x1000.jpg);"&gt;
          &lt;span class="h-u-visually-hidden"&gt;Demo of Agent Executor&lt;/span&gt;
        &lt;/div&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-L5Iw1IrZ6Nc-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="L5Iw1IrZ6Nc"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=L5Iw1IrZ6Nc"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Get started today&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Models, agents, harnesses, and the infrastructure around them are all evolving faster than ever. We’re building Agent Executor in the open so we can validate the design in the hands of real developers and improve based on your feedback.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Agent Executor is available now in preview. We invite you to explore the code, test it with your own workloads, and help shape the future of agent runtimes. Head over to our &lt;/span&gt;&lt;a href="https://github.com/google/ax" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;GitHub repo&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to get started today. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 20 May 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/ai-machine-learning/agent-executor-googles-distributed-agent-runtime/</guid><category>Application Development</category><category>AI &amp; Machine Learning</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Introducing Agent Executor, Google’s distributed Agent Runtime</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/ai-machine-learning/agent-executor-googles-distributed-agent-runtime/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Jaana Dogan</name><title>Software Engineer</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Ethan Bao</name><title>Engineering Director</title><department></department><company></company></author></item><item><title>Transforming how applications are built and managed in the AI era</title><link>https://cloud.google.com/blog/products/application-development/new-app-capabilities-for-the-entire-software-lifecycle/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;AI is massively accelerating the speed at which developers create code. This velocity is shifting the bottleneck from creation to operations, requiring SREs, DevOps, and platform engineers to take that code into production at the necessary speed and scale — all while ensuring it adheres to critical compliance, security, and governance requirements. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To solve this challenge, we announced significant new capabilities and innovations across our portfolio at Google Cloud Next ‘26, including &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/application-design-center/docs/overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Application Design Center,&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/app-hub/docs"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;App Hub/App Topology/App Optimize&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, and &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/hub/docs/overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Hub&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. These new capabilities transform the entire software lifecycle, helping you turn AI-generated code into enterprise-ready applications that are secure, compliant, and ready for production scale. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In this post, we’ll deep dive into the new capabilities and how you can get started. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Accelerate development velocity with &lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt;Application Design Center&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The Application Design Center enables teams to accelerate the design and deployment of cloud infrastructure, while ensuring architectures adhere to organizational best practices and policies. It helps teams bridge the gap between developers creating applications and the complexities of cloud architecture by providing a library of customizable, pre-built, Google-recommended architecture templates that have security and governance built-in. &lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;strong style="vertical-align: baseline;"&gt;1. Based on best-practice architectural templates&lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;You can develop applications faster with Application Design Center’s growing library of opinionated application best practice architectures. Platform admins can start with pre-defined out-of-the-box architectures, and customize them to curate a collection of Application Templates for their developers to use. This provides developers a trusted, self-service experience they can quickly discover and deploy compliant applications. These templates are also fully integrated with Gemini Cloud Assist, so you can use natural language prompts, including in AI-powered IDEs. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1._ADC_in_IDE_with_Agent_Governance.max-1000x1000.png"
        
          alt="1. ADC in IDE with Agent Governance"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p role="presentation" style="padding-left: 40px;"&gt;&lt;strong style="vertical-align: baseline;"&gt;2. Fully deployable infrastructure as code&lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Application Design Center generates deployable Terraform code from your best-practice architecture templates, including &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/application-design-center/docs/manage-application-instances#create-application-revision"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;immutable template revisions&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; for auditing architectural changes. And by detecting configuration drift, Application Design Center enables teams to remediate unauthorized changes, or deploy approved updates, ensuring continuous consistency and compliance with organizational guidelines.&lt;/span&gt;&lt;/p&gt;
&lt;p role="presentation" style="padding-left: 40px;"&gt;&lt;strong style="vertical-align: baseline;"&gt;(New)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Application Design Center now supports custom components to be added to design templates by &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/application-design-center/docs/import-components"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;ingesting existing Terraform configurations&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. This allows organizations to design using their existing Terraform components. We have also added support to integrate with your preferred CI/CD pipelines, so your teams can follow their existing deployment workflows. In addition, you can now add partner solutions directly into your designs, with support for products from MongoDB, Elasticsearch, Neo4j, and Palo Alto Networks.&lt;/span&gt;&lt;/p&gt;
&lt;p role="presentation" style="padding-left: 40px;"&gt;&lt;strong style="vertical-align: baseline;"&gt;3. Understanding business semantics &lt;br/&gt;&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;The Application Design Center treats the "Application" as the central unit of management. By maintaining the application as a core resource, you may define policies or perform observability, troubleshooting, and optimizations on complete applications versus having to manage resources independently. &lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;strong style="vertical-align: baseline;"&gt;(New)&lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Application Design Center now pulls-forward security adherence at design time. This &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;includes automated governance as designs created by &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Gemini &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Cloud Assist comply with your organization’s policies, and "pre-flight checks" before Applications are deployed. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;E&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;very application follows organizational guidelines and compliance policies before it ever hits production. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-paragraph_with_image"&gt;&lt;div class="article-module h-c-page"&gt;
  &lt;div class="h-c-grid uni-paragraph-wrap"&gt;
    &lt;div class="uni-paragraph
      h-c-grid__col h-c-grid__col--8 h-c-grid__col-m--6 h-c-grid__col-l--6
      h-c-grid__col--offset-2 h-c-grid__col-m--offset-3 h-c-grid__col-l--offset-3"&gt;

      






  

    &lt;figure class="article-image--wrap-small
      
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/media_markt.max-1000x1000.jpg"
        
          alt="media markt"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  





      &lt;p data-block-key="5buia"&gt;“Enterprise engineering is not solved by AI agents writing more code; it is about the context and standards that define how we build and run software. By embedding our architectural and governance guardrails directly into IDEs and CLIs via Gemini Cloud Assist and the Application Design Center MCP, we are creating a truly agentic developer experience. Platform teams provide the framework, and developers consume it instantly within their workflow. This shifts the focus from producing code to deploying production-ready, compliant solutions from Day 0. The result is faster delivery grounded in enterprise reality.” &lt;b&gt;&lt;i&gt;- Ervis Duraj, Principal Engineer, MediaMarktSaturn Technology&lt;/i&gt;&lt;/b&gt;&lt;/p&gt;
    &lt;/div&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Connect the dots with &lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt;App Hub and App Topology&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Once you have provisioned your infrastructure using the best-practice templates in the Application Design Center, the challenge shifts to maintaining operational consistency and visibility at scale. To do this, your teams need a clear, shared understanding of all your running applications and how they connect. App Hub and App Topology provide this single source of truth.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;App Hub acts as a central catalog for all your applications and services. It automatically discovers your cloud infrastructure and organizes it in a way that makes sense for your team. This frees your engineers from having to manually track complex infrastructure. With the new&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;developer insight connect integration, see the lineage of source code and commits that impacted your application, allowing you to see the full picture of your software development lifecycle. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Building on this foundation, App Topology delivers a unified, semantic graph that makes the relationships between your cloud resources visible, understandable, and actionable. This is crucial for rapid troubleshooting and impact analysis, and serves as the foundation for AI workflows to understand how applications are constructed to provide informed, actionable insights. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/2._App_Topology_in_Cloud_Hub.gif"
        
          alt="2. App Topology in Cloud Hub"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Streamline app management with &lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt;Cloud Hub &lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Once your applications are organized in App Hub and visualized with App Topology, the next step is to manage them effectively. Cloud Hub acts as a unified command center, using AI-driven insights and smart recommendations to help you prioritize and answer the crucial question: "What requires my immediate attention?" &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The new Security &amp;amp; Compliance board provides security insights, complemented with Gemini Cloud Assist recommendations for triaging and remediating security findings.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With the new application topology board, you can map service dependencies and recognize how changes impact upstream and downstream components. App topology board now centralizes the connections between application elements and various data domains, making them actionable and clear. The new integration with Gemini Cloud Assist also enables AI-native troubleshooting to resolve issues more quickly.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Once your app is up and running, utilize the Cloud Hub Optimization board to get a detailed cost breakdown for application components and their underlying infrastructure, which now supports BigQuery and Networking costs.&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;The board offers high-level insights to help clarify shifts in deployment costs and identify their primary drivers. Furthermore, you can now access cost performance reports for all your applications through the App Optimize APIs.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/3._GCA_Optimization_high_res.gif"
        
          alt="3. GCA Optimization (high res)"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_with_image"&gt;&lt;div class="article-module h-c-page"&gt;
  &lt;div class="h-c-grid uni-paragraph-wrap"&gt;
    &lt;div class="uni-paragraph
      h-c-grid__col h-c-grid__col--8 h-c-grid__col-m--6 h-c-grid__col-l--6
      h-c-grid__col--offset-2 h-c-grid__col-m--offset-3 h-c-grid__col-l--offset-3"&gt;

      






  

    &lt;figure class="article-image--wrap-small
      
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/bbva.max-1000x1000.jpg"
        
          alt="bbva"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  





      &lt;p data-block-key="5buia"&gt;“Managing a global banking infrastructure often meant struggling with fragmented data and scattered telemetry which impacted our incident response. With Cloud Hub, we have achieved a single pane of glass that cuts through the noise and provides us global visibility into our applications. As we look forward, App Topology will empower our SRE and DevOps teams with cross-domain insights, setting a new standard for reliability and incident response.” &lt;b&gt;&lt;i&gt;– Gerardo Mongelli, Senior Cloud Architect&lt;/i&gt;&lt;/b&gt;&lt;/p&gt;
    &lt;/div&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Google Cloud integrates AI-supported development with AI-assisted intelligent operations and troubleshooting to help your teams accelerate innovation. This approach ensures compliance with organizational standards while building and maintaining secure, high-performing, and reliable applications designed to achieve your business goals.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Get started &lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To find out more about how Google Cloud can support your organization to transition to the AI era of software delivery, we invite you to try out the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/hub/docs/application-centric-google-cloud"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Application-centric Google Cloud&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;sub&gt;&lt;em&gt;&lt;span style="vertical-align: baseline;"&gt;1 and 2. HashiCorp’s The Cloud Complexity Report 2025, 52% of organizations identified the complexity of managing multi/hybrid cloud environments as their biggest infrastructure challenge hindering progress.&lt;/span&gt;&lt;/em&gt;&lt;/sub&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 14 May 2026 18:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/application-development/new-app-capabilities-for-the-entire-software-lifecycle/</guid><category>Application Development</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Transforming how applications are built and managed in the AI era</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/application-development/new-app-capabilities-for-the-entire-software-lifecycle/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Adam Michelson</name><title>Group Product Manager, Google Cloud</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Anoosh Saboori</name><title>Director, Product Management, Google Cloud</title><department></department><company></company></author></item><item><title>Day 2 at Google Cloud Next: A marathon developer keynote</title><link>https://cloud.google.com/blog/topics/google-cloud-next/next26-day-2-recap/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;At Google Cloud, every day is Developer Day, but none so much as day 2 of Google Cloud Next, when we hold the developer keynote.  This year’s topic? An in-depth look at &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise-agent-platform"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Enterprise Agent Platform&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. This year’s theme? Planning a marathon for 10,000 participants through the Las Vegas Strip.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;OK, let’s run with it. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Gemini Enterprise Agent Platform: A warm up&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As the evolution of Vertex AI, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Platform&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; “allows you to build autonomous agents that proactively help users — and complete tasks independently,” said Brad Calder, President, GCP and SRE. The platform does so with a whole suite of tools and capabilities to build, scale, govern, and optimize your agents. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_cBnVOvk.max-1000x1000.jpg"
        
          alt="1"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Brad then passed the baton to keynote emcees Richard Seroter, Chief Evangelist, and Emma Twersky, Developer Relations Engineer, for an in-depth run-through of the agentic marathon simulator.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The system uses three main agents: A planner to determine routes; an evaluator that assesses routes based on business and community requirements; and a simulator that takes the route, adding actors and randomized behaviors to test the impact on the city,&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This, Emma said, turns out to be “a great example of how agents can help us plan, simulate, and think about solving a really big challenge.” &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/2_u7NylGU.max-1000x1000.jpg"
        
          alt="2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;First off the blocks: Building the agent&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Mofi Rahman, Developer Relations Engineer, came onstage to demo how the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Development Kit (ADK)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, Google Cloud remote &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Model Context Protocol (MCP) servers&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Runtime&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; provide the planner agent with the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Instructions&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Skills&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Tools&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; it needs to improve the initial agent. By the end of the demo, the simulator had generated a route. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“The simulated route looks beautiful,” said Mofi. “Looks like the runners are going to get an amazing view of the entire Las Vegas Strip.”&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/3_Fz2OBHO.max-1000x1000.jpg"
        
          alt="3"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;An agent to evaluate the agent&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Next, Ivan Nardini, Developer Relations Engineer, and Casey West, Architecture Advocate, showed us how to evaluate the agent, and build a UI for it. “We want to show you how to move from fragile, unpredictable agentic loops, to a rigorously evaluated network of experts that literally build their own UI,” Casey said.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;They did so by deploying a separate model to judge the route, checking both deterministic (e.g., route length) and non-deterministic (e.g., community impact) criteria. For UI development, they showed off the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent-to-User Interface&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, or &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;A2UI&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, an open-source standard developed by Google that created an interface in a single shot. They also used the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent-to-Agent Protocol&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, or &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;A2A&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, and Agent Platform’s &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Registry&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, to connect and see which agents are deployed. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“Think of Agent Registry as the DNS of your internet of agents,” Casey said. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Agents that never forget&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Richard then mused about how to build agents that get better with time, that “take the learnings from the simulation and optimize for the next run.” Because the answer shouldn’t be to “cram raw text in every request we send back to our agents.” &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To capture this learned knowledge, Agent Platform offers &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Platform Sessions&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Memory Bank&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, plus the ability to turn to tools like Spark or a database to retrieve more information, resulting in an even stronger simulator.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;When agents go off course&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Thus far, everything had gone swimmingly, but then Richard accidentally “broke” the simulator agent. That provided a perfect opportunity for Megan O’Keefe, Senior Staff Developer Advocate, to show off &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Interoperability&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Gemini Cloud Assist&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, and how to use them to debug agents at scale. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“With these autonomous agents, the production challenge isn’t just scaling the infrastructure, it’s managing the reasoning, the tool calls — all the places in the system where something can go wrong!” Megan said. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Megan used &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Runtime trace view&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; to see where the problem was, and using natural language, launched a &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Cloud Assist Investigation&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; to explore logs and events, which pointed to a specific line of code as the offender. Megan then opened up her &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Antigravity IDE&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; (powered by &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Gemini 3&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, and connected via MCP) to find the problem (an insufficiently run “event compaction” run) and to suggest a fix (add a token_threshold parameter to the event compaction config). She approved the fix and committed it to source, triggering a redeployment to &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Platform&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. Problem solved!&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Scaling the agents&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To this point, all of the presenters had been showing off agent services running as &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Cloud Run&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; services. Bobby Allen, Group Product Manager, then showed how to convert the apps to &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Google Kubernetes Engine (GKE)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, which provides greater control, as well as to use a customized &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Gemma 4&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; model, all by vibe coding in the Antigravity editor, which is connected to Cloud Assist. Along the way, Bobby also migrated the agents from &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;GCSFuse&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; to a high-performance &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Lustre &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;file system. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Closely related to scaling is sharing — making agents available for the world to use and build on. Ines Envid, Senior Director, Product Management and Jason Davenport, Area Technical Lead, showed how to build no-code agents from the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Gemini Enterprise&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; app, and how to integrate them with other, “high-code” agents. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Shifting down&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Last but not least, it was time to talk about security and governance. “Agents give users and other agents new ways to intentionally — or unintentionally — expose data and behavior in ways that we may not want,” mused Emma. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The standard response to that is to “shift left” — move testing, quality, and performance evaluation earlier in the development process — but for developers, that usually means more work, Richard said. “It’s not sustainable for developers to be responsible for all the layers of the stack,” he said. Instead, “we need to shift down.”&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To help, there’s &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Identity&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Gateway&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, demoed by Ankur Kotwal, head of Cloud Developer Relations. Ankur showed how Agent Gateway uses IAM policies to ensure agent actions are only accessible by approved sources, and how Agent Identity provides each agent with a unique and immutable credential. Then, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Policies&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; can be configured to provide guardrails for the agents.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Yinon Costica, Co-Founder and VP of Product at Wiz, then went a step further and showed how Wiz can scan your agent code and infrastructure, and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Wiz Green Agent&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; can suggest root cause remediations. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“It’s a full architecture for security to easily understand what you built without you having to actually explain it,” Yinon said. Better yet, he also showed using this functionality from &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Anthropic’s Claude Code with Opus. &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;“With Wiz, we want to enable your choice of tools and models to fix and prevent real risks,” he said.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=A01DQ8_xy7Q"
      data-glue-modal-trigger="uni-modal-A01DQ8_xy7Q-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        

        &lt;div class="article-video__aspect-image"
          style="background-image: url(https://storage.googleapis.com/gweb-cloudblog-publish/images/maxresdefault_Mz6wY0B.max-1000x1000.jpg);"&gt;
          &lt;span class="h-u-visually-hidden"&gt;Google Cloud Next &amp;#x27;26 Developer Keynote&lt;/span&gt;
        &lt;/div&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-A01DQ8_xy7Q-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="A01DQ8_xy7Q"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=A01DQ8_xy7Q"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;The finish line&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With this, the developer keynote came to an end. But for Google Cloud developers, it’s just the beginning, as the entire solution is available as source code in GitHub, and all the &lt;/span&gt;&lt;a href="https://codelabs.developers.google.com/next26/dev-keynote/build-multi-agent-marathon-planner#0" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;demos are available as Codelabs&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. Because when it comes to agentic development, these resources will really help you hit the ground running. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 23 Apr 2026 13:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/google-cloud-next/next26-day-2-recap/</guid><category>AI &amp; Machine Learning</category><category>Application Development</category><category>Google Cloud Next</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0423_102940-5172_ALIVE.max-600x600.jpg" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Day 2 at Google Cloud Next: A marathon developer keynote</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0423_102940-5172_ALIVE.max-600x600.jpg</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/google-cloud-next/next26-day-2-recap/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Google Cloud Content &amp; Editorial </name><title></title><department></department><company></company></author></item><item><title>Day 1 at Google Cloud Next ‘26 recap</title><link>https://cloud.google.com/blog/topics/google-cloud-next/next26-day-1-recap/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Last year at Google Cloud Next ‘25, we asked you to imagine a new future for AI. At Next ‘26, the question before you is how do you move AI into production across your entire enterprise?&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;According to Google Cloud CEO Thomas Kurian, the answer is straightforward: You need a unified stack, with “chips that are designed for models, models that are grounded in your data, agents and applications that are built with those models,” and the whole thing “secured by the infrastructure,” Thomas said in his keynote. (This is the same unified stack that Google uses for Search, YouTube, Chrome, and Android. As Alphabet CEO Sundar Pichai said in his opening remarks, “a big focus of ours  is to always be customer zero for our own technologies.”)&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As AI matures, we’ve laid out a blueprint on how to succeed. Read on for a whirlwind tour of what we announced from the keynote stage&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=11PBno-cJ1g"
      data-glue-modal-trigger="uni-modal-11PBno-cJ1g-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        

        &lt;div class="article-video__aspect-image"
          style="background-image: url(https://storage.googleapis.com/gweb-cloudblog-publish/images/next26_live_stream.max-1000x1000.jpg);"&gt;
          &lt;span class="h-u-visually-hidden"&gt;Google Cloud Next &amp;#x27;26 Opening Keynote&lt;/span&gt;
        &lt;/div&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-11PBno-cJ1g-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="11PBno-cJ1g"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=11PBno-cJ1g"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Gemini Enterprise: The end-to-end system for the agentic era&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Throughout this unified stack is Gemini Enterprise — “the connective tissue between your data, your people, and your goals,” Thomas said, providing a combination of intelligence and automation across multiple layers. Here’s what’s new.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0422_103840-5416_ALIVE.max-1000x1000.jpg"
        
          alt="GCNEXT2026_0422_103840-5416_ALIVE"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3 role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;1. Gemini Enterprise Agent Platform&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Gemini Enterprise Agent Platform is where you go to build, scale, govern, and optimize agents. As the evolution of Vertex AI, it’s built on top of our leading infrastructure, and deeply integrated with our data and security capabilities — the foundation of the Agentic Enterprise. Here’s a sampling of Agent Platform’s new features and capabilities:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Build:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Choose the right environment for the job — from the low-code, visual interface of the new &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Studio&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, to the code-first logic of the upgraded &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Development Kit (ADK)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. We’ve simplified the entire lifecycle with AI-native coding capabilities to help you ship production-grade agents faster.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Scale:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Clear the path to production with the re-engineered &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Runtime&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. This supports long-running agents that maintain state for days at a time and are backed by &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Memory Bank&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; for persistent, long-term context.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Govern:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Establish centralized control with &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Identity&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Registry&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Gateway&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. These capabilities help ensure every agent — whether built on Agent Platform or sourced from our partner ecosystem — has a trackable identity and operates within enterprise-grade guardrails. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Optimize:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Guarantee quality with &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Simulation&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Evaluation&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Observability&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. These tools provide full execution traces and a real-time lens into agent reasoning to help ensure your agents always hit their goals.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To dive deep into Agent Platform, read more in our announcement blog &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise-agent-platform"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;here&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/p&gt;
&lt;h3 role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;2. Gemini Enterprise app&lt;/span&gt;&lt;/h3&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0422_091032-8701_ALIVE.max-1000x1000.jpg"
        
          alt="GCNEXT2026_0422_091032-8701_ALIVE"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The Gemini Enterprise app is&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; “the primary environment where your business actually operates,” Thomas explained. The app is where many workers, especially non-technical ones, can ask questions of enterprise agents, create generative media, engage with prebuilt agents, and even create their own with conversational interfaces — all with governance, compliance, and security built in. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Here’s a sample of what’s new in this foundational interface: &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Gemini Enterprise Projects &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;give your agents permanent memory.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Deep Think &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;solves your most complex business challenges without context pollution.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Microsoft 365 &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;interoperability makes it easy to export docs you create with Canvas into Microsoft Office formats.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To illustrate the power of Gemini Enterprise, Shaun White, three-time Olympic gold medalist, entrepreneur, and snowboarding legend, joined us on stage.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“Back when I was training, our tools were camcorders and guesswork. You’d land a trick and watch it back. And you’d be thinking, ‘How can I make that trick better?’” he said.  &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Alongside Jason Davenport, Google Cloud Tech Lead, they showed a model that Google Cloud built in collaboration with Google DeepMind that tracked Shaun in space from a two-dimensional video, helping him understand what he was doing right and wrong. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“Learning the trick on the mountain is one thing, but actually understanding the physics of a trick is a whole other thing,” Shaun said.  &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Read more on the Gemini Enterprise app &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/whats-new-in-gemini-enterprise?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;here&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0422_093546-4510_ALIVE.max-1000x1000.jpg"
        
          alt="Shaun White"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;div data-draftjs-conductor-fragment='{"blocks":[{"key":"dsr0p","text":"3. AI Hypercomputer","type":"unstyled","depth":0,"inlineStyleRanges":[],"entityRanges":[],"data":{}},{"key":"4v39t","text":"The same technology foundation that athletes like Shaun White use to understand their performance is being used by enterprises to transform their businesses. ","type":"unstyled","depth":0,"inlineStyleRanges":[],"entityRanges":[],"data":{}},{"key":"fhjv0","text":"Amin Vadhat, SVP and chief technologist, AI and Infrastructure, took to the stage to announce enhancements to AI Hypercomputer, the integrated supercomputing underneath every AI workload on Google Cloud.","type":"unstyled","depth":0,"inlineStyleRanges":[],"entityRanges":[{"offset":110,"length":16,"key":0}],"data":{}}],"entityMap":{"0":{"type":"LINK","mutability":"MUTABLE","data":{"url":"https://cloud.google.com/solutions/ai-hypercomputer"}}}}'&gt;
&lt;div class="Draftail-block--unstyled" data-block="true" data-editor="fujua" data-offset-key="ma2lh-0-0"&gt;
&lt;h3 class="public-DraftStyleDefault-block public-DraftStyleDefault-ltr" data-offset-key="ma2lh-0-0"&gt;&lt;span data-offset-key="ma2lh-0-0"&gt;3. AI Hypercomputer&lt;/span&gt;&lt;/h3&gt;
&lt;/div&gt;
&lt;div class="Draftail-block--unstyled" data-block="true" data-editor="fujua" data-offset-key="25o56-0-0"&gt;
&lt;div class="public-DraftStyleDefault-block public-DraftStyleDefault-ltr" data-offset-key="25o56-0-0"&gt;&lt;span data-offset-key="25o56-0-0"&gt;The same technology foundation that athletes like Shaun White use to understand their performance is being used by enterprises to transform their businesses. &lt;/span&gt;&lt;/div&gt;
&lt;/div&gt;
&lt;div class="Draftail-block--unstyled" data-block="true" data-editor="fujua" data-offset-key="et8mb-0-0"&gt;
&lt;div class="public-DraftStyleDefault-block public-DraftStyleDefault-ltr" data-offset-key="et8mb-0-0"&gt; &lt;/div&gt;
&lt;div class="public-DraftStyleDefault-block public-DraftStyleDefault-ltr" data-offset-key="et8mb-0-0"&gt;&lt;span data-offset-key="et8mb-0-0"&gt;Amin Vadhat, SVP and chief technologist, AI and Infrastructure, took to the stage to announce enhancements to &lt;/span&gt;&lt;a class="TooltipEntity" data-draftail-trigger="true" href="https://cloud.google.com/solutions/ai-hypercomputer" role="button"&gt;&lt;span data-offset-key="et8mb-1-0"&gt;AI Hypercomputer&lt;/span&gt;&lt;/a&gt;&lt;span data-offset-key="et8mb-2-0"&gt;, the integrated supercomputing underneath every AI workload on Google Cloud.&lt;/span&gt;&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0422_094223-0331_ALIVE.max-1000x1000.jpg"
        
          alt="GCNEXT2026_0422_094223-0331_ALIVE"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;First, there’s the eighth-generation Tensor Processing Unit, or TPU — “a thing of beauty,” Amin said. And because “the demands of training and serving have completely diverged,” this TPU family actually consists of two chips: &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;TPU 8t, optimized for training&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, uses new Inter-Chip Interconnect (ICI) technology to scale up to 9,600 TPUs and 2 petabytes of shared, high-bandwidth memory in a single superpod. It achieves three times the processing power of Ironwood and delivers up to double the performance per watt. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;TPU 8i, optimized for inference&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, uses the new Boardfly topology to directly connect 1,152 TPUs in a single pod. It features three times more on-chip SRAM compared to previous versions and a specialized Collectives Acceleration Engine offloads resource-heavy tasks. Taken together, TPU 8i delivers 80% better performance per dollar for inference than the prior generation, helping &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/compute/tpu-8t-and-tpu-8i-technical-deep-dive"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;millions of concurrent agents to run cost-effectively&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;AI Hypercomputer also supports Arm-based Google Cloud Axion processors, such as the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;N4A, now generally available&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, and will be one of the first platforms to offer &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;NVIDIA’s Vera Rubin NVL72&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; platform when it is released.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Other parts of the network need to keep up with the demands of the agentic era. For example, the new &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Virgo Network&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; doubles connectivity to scale training beyond AI Hypercomputer superpods, and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Google Cloud Managed Lustre&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; now supports an industry-leading 10 terabytes per second of throughput. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Learn more about all of our AI infrastructure innovations &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/compute/ai-infrastructure-at-next26?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;here&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/p&gt;
&lt;h3 role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;4. Agentic Data Cloud&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The AI era hinges on data. Lots of data. That data comes with a catch: It needs to be grounded in context. That’s because “reasoning without context is just a guess,” explained Karthik Narain, chief product and business officer.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0422_085702-_ALIVE.max-1000x1000.jpg"
        
          alt="GCNEXT2026_0422_085702-_ALIVE"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To that end, we’re totally rethinking our data platform, and giving it a new name: the Agentic Data Cloud.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Here’s a sampling of what you’ll find in the Agentic Data Cloud:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Knowledge Catalog&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; constructs a unified, dynamic context graph of your entire business enabling you to ground agents in all of your business data and semantics. With &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Smart Storage&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; and the Object Context API, files in Google Cloud Storage are instantly tagged and enriched with metadata before an agent touches them. Knowledge Catalog is also integrated with Gemini’s &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Deep Research Agent&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Data Agent Kit&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; delivers a Gemini-powered data science authoring experience across your IDEs, Notebooks, and agentic terminals. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Lightning Engine for Apache Spark &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;is a real-time, serverless engine that is up to 4.5 times faster than open-source alternatives and offers up to double price-performance over the leading competitor for large datasets.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Finally, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Cross-Cloud Lakehouse&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, based on Apache Iceberg, lets you query data in Amazon Web Services or Azure without having to copy it.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Learn all about all the innovations in the Agentic Data Cloud &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/data-analytics/whats-new-in-the-agentic-data-cloud?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;here&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;5. Agentic Defense&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;AI makes — and demands — that everything go faster, and security operations are no exception. Increasingly, “human analysts can’t keep up with AI-driven attacks,” said Francis deSouza, COO, and president, Security Products. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“Security must become an autonomous force, responding faster than the threat itself,” he said. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To help, we introduced three new agents in &lt;/span&gt;&lt;a href="https://cloud.google.com/security/products/security-operations"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Security Operations&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to help you defend at the speed of AI. &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Threat Hunting agent&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Helps teams proactively hunt for novel attack patterns and stealthy adversary behaviors that bypass traditional defenses. Now in preview. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Detection Engineering agent&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Identifies coverage gaps and creates new detections for threat scenarios. Now in preview. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Third-Party Context agent: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Enriches workflows with contextual data from third-party content. Coming soon to preview.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;You can also &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;build your own security agents&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; with remote Google Cloud MCP server support for Google Security Operations, now generally available, and access it from the Google Security Operations chat interface, now in preview.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0422_100749-9822_ALIVE.max-1000x1000.jpg"
        
          alt="GCNEXT2026_0422_100749-9822_ALIVE"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Then there’s &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/identity-security/google-completes-acquisition-of-wiz?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Wiz, now a part of Google Cloud&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, whose AI-Application Protection Platform (AI-APP), Wiz Security Agents, and Wiz Workflow help you identify and respond to risks and threats at machine speed. New in the Wiz family today are&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;:  &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Secure vibe-coded applications: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;A new integration runs Wiz security scanning directly inside the Lovable platform so vulnerabilities, secrets, and misconfigurations caught by Wiz surface in Lovable's built-in security view. Generally available in May.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Secure AI-generated code&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Wiz can now remove risks from AI-generated code with inline AI security hooks integrated directly into IDEs and agent workflows, injecting security guardrails before code is committed.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent-based remediation&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Wiz Skills can equip coding agents and AI-native IDEs with full code-to-cloud context and validated attack surface findings from the Wiz Security Graph.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://www.wiz.io/academy/ai-security/ai-bom-ai-bill-of-materials" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;AI-Bill of Materials&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt; (AI-BOM):&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Work towards eliminating shadow IT by automatically inventorying all AI frameworks, models, and IDE extensions across your environment.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://cloud.google.com/blog/products/identity-security/introducing-google-cloud-fraud-defense-the-next-evolution-of-recaptcha"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud Fraud Defense&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;: The evolution of reCAPTCHA, this platform is designed to discern the legitimacy and authorization of bots, humans, and agents. Now generally available.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Read more about these security innovations &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/identity-security/next26-redefining-security-for-the-ai-era-with-google-cloud-and-wiz?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;here&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/p&gt;
&lt;h3 role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;6. Workspace Intelligence&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For a look at AI for end-users, we heard from Yulie Kwon Kim, VP, Product, Google Workspace, who shared new ways that AI is manifesting in our collaboration and productivity suite. &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Workspace Intelligence &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;is a unifying semantic layer that breaks down information and context silos for you and your agents. It understands your work, your priorities and the people you work with to help you get more done. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0422_102731-1098_ALIVE.max-1000x1000.jpg"
        
          alt="GCNEXT2026_0422_102731-1098_ALIVE"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“Think of it as a unified intelligence layer that lives inside every Workspace app. It connects the dots and lets AI do the heavy lifting," Yulie said.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Here’s what’s new:&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Ask Gemini in Google Chat &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;allows you to instantly synthesize information, surface insights, and query projects from across Workspace directly from your Google Chat window. It provides proactive daily briefings to help you prioritize, and also lets you take immediate action — such as scheduling a meeting on your calendar or creating a Google Doc to develop a pre-meeting brief — turning your conversations into momentum without having to switch tabs.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=YppfLqH7Fps"
      data-glue-modal-trigger="uni-modal-YppfLqH7Fps-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        

        &lt;div class="article-video__aspect-image"
          style="background-image: url(https://storage.googleapis.com/gweb-cloudblog-publish/images/maxresdefault_JTY9905.max-1000x1000.jpg);"&gt;
          &lt;span class="h-u-visually-hidden"&gt;Ask Gemini in Chat&lt;/span&gt;
        &lt;/div&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-YppfLqH7Fps-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="YppfLqH7Fps"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=YppfLqH7Fps"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Reimagined content creation in Docs, Sheets, and Slides &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;uses Workspace Intelligence to synthesize information from across Workspace and the web and creates professionally formatted drafts that match your voice, style, and brand. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;AI Inbox and AI Overviews in Gmail&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; creates a&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; personal, proactive &lt;/span&gt;&lt;a href="https://blog.google/products-and-platforms/products/gmail/gmail-is-entering-the-gemini-era/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;inbox assistant with Gemini&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Google Drive Projects&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; instantly organizes your team's files and emails to manage workflows, generate content, and deliver specific answers based on rich project context. In addition to newly added &lt;/span&gt;&lt;a href="https://blog.google/products-and-platforms/products/workspace/gemini-workspace-updates-march-2026/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;AI Overviews and Ask Gemini&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, Projects is another way we’re transforming Drive from a storage tool into an active collaborator to provide insights about your data.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Workspace agent&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; in Gemini Enterprise executes complex, multi-step tasks across Google Workspace apps without having to leave Gemini Enterprise. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For more, check out the full Workspace Intelligence &lt;/span&gt;&lt;a href="https://workspace.google.com/blog/product-announcements/introducing-workspace-intelligence?_gl=1*1lcx1ks*_up*MQ..&amp;amp;gclid=CjwKCAjw46HPBhAMEiwASZpLRCh04El-PH-mQX3OW7IcONinrI6ZdqmWKi_j1tyhxEOFnZTaaMBr2xoCFb8QAvD_BwE&amp;amp;gclsrc=aw.ds" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;blog&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3 role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;7. Agentic Commerce&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For enterprises, AI agents are reshaping how consumers engage with companies and their products. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“In the agentic era, an agent isn’t just a tool; it’s a strategic extension of your business, built to expand your reach, deepen engagement, and personalize service at scale,” said Carrie Tharp, Google Cloud Vice President of Go To Market Strategic Industries. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href="https://cloud.google.com/products/gemini-enterprise-for-customer-experience"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Enterprise for Customer Experience&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; offers a suite of tools to enhance the entire customer journey, from the first moment of discovery through on-going interactions that remember the customer like the best shopkeeper would. &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Shopping agent &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Food Ordering agent &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;bring new conversational sales and ordering capabilities direct to businesses and third-party chat and digital interfaces.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Omnichannel Gateway&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; helps agents maintain context across web, mobile, and voice, so a company’s agents can offer more personalized service.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Assist&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; helps during complex customer service situations, coaching employees to deliver fast and more accurate answers to customer questions by having organizational data readily available through gen AI grounding.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With Omnichannel Gateway in particular, about bridging the physical, digital, and agentic shopping experience, so consumers always have a familiar, brand-aware experience. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“If a customer moves from text chat to a phone call, the agent seamlessly remembers exactly where they left off,” said Carrie. Now that’s progress!&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For more customer stories, check out all &lt;/span&gt;&lt;a href="https://cloud.google.com/transform/101-real-world-generative-ai-use-cases-from-industry-leaders?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;1,302 of the latest gen AI use cases&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; from businesses around the globe.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Innovate all the things &lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;From new products, to new solutions, to new ways of working, there are so many other ways that we’re helping organizations take their AI from pilot to production. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Over the following week, we’ll share even more news, helpful how-to guides, and go deeper on today’s announcements. Stay tuned!&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 22 Apr 2026 23:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/google-cloud-next/next26-day-1-recap/</guid><category>AI &amp; Machine Learning</category><category>Application Development</category><category>Data Analytics</category><category>Google Cloud Next</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0422_090309-3826_ALIVE.max-600x600.jpg" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Day 1 at Google Cloud Next ‘26 recap</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/GCNEXT2026_0422_090309-3826_ALIVE.max-600x600.jpg</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/google-cloud-next/next26-day-1-recap/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Google Cloud Content &amp; Editorial </name><title></title><department></department><company></company></author></item><item><title>What’s new in GKE at Next ‘26</title><link>https://cloud.google.com/blog/products/containers-kubernetes/whats-new-in-gke-at-next26/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This week at Google Cloud Next ‘26, we are sharing the evolution of Google Kubernetes Engine (GKE), delivering leading performance, efficiency, security, and scale for your most demanding and complex workloads, and the next generation of AI and agentic applications.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Why it matters:  &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Kubernetes has rapidly become the operating system for the AI era, with &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;GKE now powering AI workloads for all of our top 50&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; customers on the platform, including the largest frontier model builders.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; We are witnessing a massive acceleration in enterprise AI. In just a few months, the number of &lt;/span&gt;&lt;a href="https://www.databricks.com/blog/enterprise-ai-agent-trends-top-use-cases-governance-evaluations-and-more" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;multi-agent AI workflows has surged&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; by 327%. At the same time, &lt;/span&gt;&lt;a href="https://thenewstack.io/cncf-kubernetes-is-foundational-infrastructure-for-ai/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;66%&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; of organizations rely on Kubernetes to power generative AI apps and agents.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This new era of autonomous agents operating at massive scale requires a foundational change in how we manage infrastructure — a change that is more demanding than the shift from stateless to stateful applications. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;What’s new: &lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;GKE Agent Sandbox:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Secure, highly scalable, low-latency agent infrastructure&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;GKE hypercluster:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;  A single, conformant GKE control plane to manage millions of accelerators across Google Cloud regions&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Improved inference performance:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Foundational enhancements to GKE Inference Gateway and KV Cache management&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Reinforcement learning (RL) enhancers: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Native capabilities to relieve bottlenecks that throttle accelerator utilization &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Scaling on custom metrics:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Support for intent-based autoscaling on triggers besides CPU and memory&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Read on for details about these GKE announcements.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;GKE Agent Sandbox: Accelerating the agentic era&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;AI evolves from simple conversational chatbots to entire ecosystems of proactive, autonomous agents, the underlying infrastructure must adapt to handle hundreds or thousands of agents collaborating with workers to plan, evaluate, and execute complex tasks. At scale, infrastructure performance, responsiveness, and rigorous security are essential. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We are excited to announce &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/kubernetes-engine/docs/concepts/machine-learning/agent-sandbox"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;GKE Agent Sandbox&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, the industry’s most scalable and low-latency agent infrastructure. Built with gVisor kernel isolation — the same technology securing Gemini — Agent Sandbox allows you to safely execute untrusted code, tools, and entire agents without sacrificing performance. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;GKE provides leading speed and efficiency for fully isolated agents with &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;300 sandboxes per second at sub-second latency and up to 30% better price-performance when running on Axion compared to other hyperscale clouds.&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Lovable empowers anyone to build apps and websites — with builders creating 200,000+ new projects daily.  Lovable runs these AI-generated applications in GKE Agent Sandboxes because of the fast startup, fast scaling and secure isolation. &lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;GKE's cutting-edge sandboxing capabilities allow us to reliably scale to hundreds of secure sandboxes per second, ensuring we can seamlessly empower builders, even during massive, unpredictable demand." &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;- Fabian Hedin, Co-founder, Lovable &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;GKE hypercluster redefines the scalability ceiling &lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As foundational AI models grow exponentially and accelerators remain in high demand, organizations resort to fracturing Kubernetes compute infrastructure into hundreds of disconnected clusters, which can create a massive operational burden. To help, we’re announcing the private GA of &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;GKE hypercluster&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, which allows a single, Kubernetes conformant GKE control plane to manage a million chips distributed across 256,000 nodes — spanning multiple Google Cloud regions. With the GKE hypercluster, widely distributed infrastructure becomes a single, unified capacity reserve that spans geographical locations.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To scale globally without compromising security, GKE hypercluster relies on Google’s Titanium Intelligence Enclave, a software-hardened security engine that delivers private AI compute. This "no-admin-access" model provides hardware-attested, pod-level isolation, so that proprietary model weights and prompts remain cryptographically sealed from platform administrators and infrastructure layers.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Supercharging state-of-the-art inference&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Achieving frontier inference requires months of complex performance tuning. To reduce this heavy lifting, GKE now slashes your "time to SOTA" across TPUs and GPUs to mere minutes. We do this with new capabilities:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;ML-driven &lt;/span&gt;&lt;a href="https://llm-d.ai/blog/predicted-latency-based-scheduling-for-llms" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Predictive Latency Boost&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in GKE Inference Gateway, which can reduce time-to-first-token latency by up 70% by replacing heuristic guesswork with real-time capacity-aware routing — no manual tuning required. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Automatic KV Cache storage tiering&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; across RAM, Local SSD, and GCS/Lustre solves long-context memory bottlenecks. &lt;/span&gt;&lt;a href="https://github.com/llm-d/llm-d/blob/main/guides/tiered-prefix-cache/README.md" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Offloading KV Cache&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to RAM yielded a more than 40% TTFT reduction and a 50% throughput gain for a 10K system prompt length. Offloading KV Cache to Local SSD yielded an almost 70% throughput improvement for a 50K system prompt length. Learn more about these benchmarks in the &lt;/span&gt;&lt;a href="https://github.com/llm-d/llm-d/blob/main/guides/tiered-prefix-cache/storage/README.md#benchmarking" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;llm-d Offloading Prefix Cache to Shared Storage guide&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Built as part of a layered composable suite, these new GKE capabilities leverage llm-d, now an official &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/containers-kubernetes/llm-d-officially-a-cncf-sandbox-project"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;CNCF Sandbox project&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. To give you maximum flexibility, we’ve partnered closely with NVIDIA to seamlessly integrate Dynamo for scaling massive &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/compute/scaling-moe-inference-with-nvidia-dynamo-on-google-cloud-a4x?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Mixture-of-Experts (MoE) models&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. Whichever tools you choose, GKE provides the highly-optimized, flexible infrastructure you need to safely run any frontier AI workload — including the advanced agentic capabilities of the newly announced &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/gemma-4-available-on-google-cloud?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemma 4&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Eliminating RL compute bottlenecks&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Reinforcement learning (RL) is a key driver of AI compute demand and RL jobs involve sequential processing for sampling, reward, and training that can leave GPU and TPU accelerators idle between these RL steps. To streamline RL, we are adding new GKE capabilities in preview:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://github.com/llm-d-incubation/py-inference-scheduler" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;RL Scheduler&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; solves for the "straggler effect" and inter-batch tail latency, maximizing throughput via intelligent routing.  &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;RL Sandbox&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; provides kernel-level isolation for tool-calling and reward evaluation with millisecond-scale provisioning. Easy integration with RL sampling and reward steps.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://docs.cloud.google.com/kubernetes-engine/docs/tutorials/monitor-reinforcement-learning-workloads"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;RL Observability and Reliability&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; dashboards offer the deep visibility required to troubleshoot and optimize the entire RL loop instantly, out of the box.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Review the RL on GKE recipe, specifically the implementations for &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/kubernetes-engine/docs/tutorials/scaling-rl-verl-gke"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Verl&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/kubernetes-engine/docs/tutorials/nemo-rl-gke"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;NeMo RL&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Intent-based autoscaling on custom metrics&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Traditionally, scaling AI workloads based on application health has imposed a "custom metric tax." To scale the system on anything but basic compute or memory utilization, organizations have to manage complex monitoring systems and IAM roles. This creates operational risk: if your external observability stack fails, your autoscaling breaks along with it.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Intent-based autoscaling&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; eliminates this overhead via native &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/kubernetes-engine/docs/how-to/expose-custom-metrics-autoscaling"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;custom metrics support&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt; for GKE’s Horizontal Pod Autoscaler (HPA)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. This agentless architecture bypasses external dependencies by sourcing metrics directly from Pods, hardening reliability while cutting costs. Crucially, it drops reaction times from 25 seconds to just 5 seconds—a &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;5x &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; performance gain for near-instantaneous infrastructure elasticity.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;New workloads, same mission&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For over a decade, GKE has set the standard for scalable infrastructure. As we enter the era of agentic and autonomous AI, our mission remains the same: eliminating operational friction so you can focus on innovation. The capabilities we are announcing at Next ‘26 — from GKE hypercluster and the Agent Sandbox, to ultra-fast inference and intent-based autoscaling — give you the secure, efficient, and powerful engine you need to succeed with your ambitious AI workloads. To learn more about using GKE for your AI workloads, check out &lt;/span&gt;&lt;a href="https://cloud.google.com/kubernetes-engine/docs/how-to/machine-learning/inference/inference-quickstart"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;GKE Inference Quickstart&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 22 Apr 2026 12:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/containers-kubernetes/whats-new-in-gke-at-next26/</guid><category>AI &amp; Machine Learning</category><category>Application Development</category><category>GKE</category><category>Google Cloud Next</category><category>Containers &amp; Kubernetes</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCN26_102_BlogHeader_2436x1200_Opt_13_Dark.max-600x600.jpg" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>What’s new in GKE at Next ‘26</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/GCN26_102_BlogHeader_2436x1200_Opt_13_Dark.max-600x600.jpg</image><site_name>Google</site_name><url>https://cloud.google.com/blog/products/containers-kubernetes/whats-new-in-gke-at-next26/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Drew Bradstock</name><title>VP Product Management, Orchestration &amp; Kubernetes</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Gari Singh</name><title>GKE Group Product Manager</title><department></department><company></company></author></item><item><title>Gemini Cloud Assist: Proactive cloud operations that work for you, even before you ask</title><link>https://cloud.google.com/blog/products/application-development/gemini-cloud-assist-at-next26/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Today at Google Cloud Next, we are unveiling a more proactive Gemini Cloud Assist, our AI-assisted cloud operations platform. This update shifts your Google Cloud operations from manual workflows to a proactive, intelligent experience supported by a powerful ecosystem of agents.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Why it matters: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;A new agentic architecture enables Gemini Cloud Assist to handle the heavy lifting of your cloud management. By embedding intelligence, your enterprise context, and the power of Gemini directly into the operational layer, Gemini Cloud Assist proactively executes complex tasks such as designing applications, troubleshooting issues, and preemptively optimizing costs, that previously required constant human oversight. In enterprise-scale systems, this approach accelerates development velocity and reduces resolution times. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;What’s new: &lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Using natural language and the power of Gemini, reduce the time from design to deployment of new or existing multi-resource deployments via a &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;redesigned Application Design Center&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Automate infrastructure operations via &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;gcloud, kubectl, and Terraform &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;while using proactive multi-turn agents to troubleshoot and resolve incidents.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Identify your cost anomalies 24/7 &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;using a proactive FinOps agent that analyzes spending spikes and generates granular cost reports on demand.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Assistance wherever you are. &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Powered by &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/mcp/supported-products"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud MCP servers&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and our proactive agents under the hood, Gemini Cloud Assist also exposes its own design, operation, troubleshooting and optimization capabilities as published MCP servers, bringing them straight to your IDE. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;“Gemini Cloud Assist has significantly helped our dev teams. It reduced the number of outreach and touch points I have with them regarding Google Cloud questions by 60%. This allows our cloud team to scale more effectively and focus on more complex tasks.” &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;- Oscar Aldana Assad, Senior Cloud Engineer, Petco&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Let’s take a deeper look at how the agentic Gemini Cloud Assist can help your operations.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Accelerate production-readiness with Application Design Center&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Gemini Cloud Assist serves as the intelligent reasoning engine for Application Design Center, acting as the bridge between natural-language intent, and a visual, production-ready architecture. By describing your infrastructure goals in plain language, Gemini Cloud Assist leverages Application Design Center to automatically lay out a visual design, including deployable Terraform. These templates are based on best-practice architecture guidance from Google Cloud and help bring security, reliability and compliance by design. Integrated with Security Command Center, quickly go from idea to deployment that conforms to your organizational policies.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Platform teams can then curate shared catalogs of pre-approved templates and integrate their own custom Terraform modules directly into the design process, providing a governed framework. This established, well-lit path helps developers adhere to organizational security and compliance guardrails from the first day of deployment. Beyond initial deployment, Gemini supports the full application lifecycle with interactive, multi-turn problem solving to update cloud resources. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Move from reactive to proactive remediation&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In production, Gemini Cloud Assist helps you shift operations from reactive troubleshooting to quickly analyzing hypotheses to drive a lower time to resolution. Triggered by alerts, Gemini Cloud Assist proactively clusters and analyzes signals to initiate investigations before issues escalate. Now with Gemini 3, Gemini Cloud Assist correlates logs and metrics and identifies root causes from infrastructure signals down to the application code. Gemini Cloud Assist explores parallel hypotheses via tool calls and presents a technical breakdown of observations in a centralized UI. If intervention is required to address an underlying Google Cloud issue, users can hand off complete context to Google support, minimizing the iterations required for sharing configuration and context data.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/proactive_alert_investigations_blog.gif"
        
          alt="proactive_alert_investigations"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Identify cost anomalies 24/7&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To maintain economic health, Gemini Cloud Assist now acts as an proactive optimization agent for your projects. Running in the background 24/7, it monitors for cost anomalies and provides root-cause analysis, correlating spending spikes with specific engineering triggers like new resource creation, auto-scaling events, or pricing changes. You can query resource utilization via natural language to generate on-demand, tabulated reports, by project and applications registered in AppHub, providing granular visibility into "&lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;who, what, when, and how&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;" — without manual data aggregation. For example, you can ask ‘Why did the cost of my application increase yesterday?’ or ‘How much did my project cost last month?’ and Gemini Cloud Assist answers by correlating cost data with infrastructure change, audit, and monitoring logs to get you an accurate answer.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/3_Vz6FwaI.gif"
        
          alt="3"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Assistance everywhere&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We are meeting teams where they work by expanding the surfaces where Gemini Cloud Assist is available. A Gemini Cloud Assist agent is already accessible through the console and mobile interfaces. And with new support for the Model Context Protocol (MCP), Gemini Cloud Assist is now available in Gemini CLI, your favorite agentic IDE or CLI, and third-party toolchains like ServiceNow and Slack. Integrating proactive assistance within existing workflows helps teams to avoid context switching and maintain flow-state.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Proactive capabilities at your fingertips&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We designed Gemini Cloud Assist to help manage the end-to-end lifecycle of your applications, providing a multi-agent approach from deploying new applications to managing existing applications in the cloud. With the help of Gemini 3, Gemini Cloud Assist can now:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Increase your development velocity: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Accelerate production-readiness using intent-driven architectures that unify best practices, security policies, and enterprise compliance.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Streamline production operations&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Triage, diagnose and resolve production issues faster, through Gemini-based troubleshooting, recommendations and remediations.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Automate cost optimization:  &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Automatically detect, analyze, root-cause, and alert you about cost anomalies for your projects on a daily basis. &lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Meet your teams where they are:  &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Through proactive agents and MCP tools, engage with functionality using surfaces that range from the Google Cloud console to your CLI and IDE, so teams can stay in a flow-state.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The future of operations is agentic. You can begin your journey with our proactive cloud by enabling &lt;/span&gt;&lt;a href="https://console.cloud.google.com/gemini-admin/products"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Cloud Assist&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in your project settings today.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 22 Apr 2026 12:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/application-development/gemini-cloud-assist-at-next26/</guid><category>AI &amp; Machine Learning</category><category>Management Tools</category><category>DevOps &amp; SRE</category><category>Google Cloud Next</category><category>Application Development</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/GCN26_102_BlogHeader_2436x1200_Opt_11_Dark.max-600x600.jpg" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Gemini Cloud Assist: Proactive cloud operations that work for you, even before you ask</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/GCN26_102_BlogHeader_2436x1200_Opt_11_Dark.max-600x600.jpg</image><site_name>Google</site_name><url>https://cloud.google.com/blog/products/application-development/gemini-cloud-assist-at-next26/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Michael Bachman</name><title>VP/GM, Cloud Foundations</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Ines Envid</name><title>Sr. Director, Product Management</title><department></department><company></company></author></item><item><title>How to build production-ready AI agents with Google-managed MCP servers</title><link>https://cloud.google.com/blog/products/ai-machine-learning/how-to-build-ai-agents-with-google-managed-mcp-servers/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As ​​developers build AI agents with more sophisticated reasoning systems, they require higher-quality fuel–in the form of enterprise data and specialized tools–to drive real business value. To get the most out of that octane-rich mix, we offer Google-managed model context protocol (MCP) servers:  an engine purpose-built for AI agents to interact securely with Google and Google Cloud services.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;These &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/mcp/overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google-hosted, fully-managed endpoints&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; allow AI agents to communicate with Google Maps, BigQuery, Google Kubernetes Engine, Cloud Run, and many other Google services. As we boldly build AI agents, ensuring that we’re also building responsibly is critical.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In this guide, we demonstrate how to build agents securely on our managed MCP servers.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Why you should use Google-managed MCP servers&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Transitioning from local experimentation to enterprise-grade AI requires adopting a robust, managed infrastructure that prioritizes scale and oversight. These are the key benefits that we offer: &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Production readiness&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: While open-source MCP servers are great for local development, they struggle in production with scalability, single points of failure, and management overhead. Google’s managed MCP servers require no infrastructure provisioning because we handle the hosting, scaling, and security.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Unified discoverability&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: You can publicly query and easily discover all available MCP endpoints for Google services (such as maps.googleapis.com/mcp) using a simple directory service.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Enterprise security&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Google MCP servers offer native integrations with the Google Cloud security stack, including Cloud IAM, VPC-SC and Model Armor.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Integrated observability and auditability&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Google MCP servers are integrated with Cloud Audit Logs, offering a centralized view of all tool-calling activity. This allows platform teams to monitor agent performance, ensure compliance, and troubleshoot interactions through a single enterprise-grade logging pane.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/Figure_1_MCP_blog.max-1000x1000.png"
        
          alt="Figure 1 MCP blog"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="sl61m"&gt;Figure 1: Google MCP Servers high-level architecture diagram&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;An AI agent example using Google MCP server with ADK&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Cityscape is a &lt;/span&gt;&lt;a href="https://github.com/danistrebel/adk-cityscape" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;demo agent&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; built with Google's Application Development Kit (ADK) that turns a simple text prompt — like "Generate a cityscape for Kyoto" — into a unique, AI-generated city image. It uses the Google Maps Grounding Lite-managed MCP server for trusted location information and the Nano Banana model (via a local MCP server) for image generation. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The lightweight app is then easily deployed to Google &lt;/span&gt;&lt;a href="https://cloud.google.com/run"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Run&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a serverless runtime, to interact with users. Below are two examples of the images generated by the agent based on the local real-time weather conditions.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/MCP_blog_figure.max-1000x1000.jpg"
        
          alt="MCP blog figure"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="sl61m"&gt;Figure 2: Example images generated by the Cityscape agent with real time weather info&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3 role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;1. Calling a Google MCP server from the ADK agent: &lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As demonstrated in the &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;get_weather&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; code snippet below, the Cityscape agent utilizes a Streamable HTTP endpoint to interface with the Google Maps MCP server. It provides the agent with real-time weather conditions for a given city, which are then used to set the atmospheric mood in the generated cityscape image. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Because it's a Google-managed remote MCP server, Google handles the hosting, scaling, and security — so your agent benefits from automatic scaling to handle any traffic level, built-in reliability with Google's production infrastructure, and enterprise-grade security out of the box. There's no infrastructure to manage — you just point to the Maps URL like below and authenticate with an API key, making it ideal for production deployments.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;# Remote Google MCP server: connects to Google Maps Grounding Lite \r\n# to fetch real-time weather conditions for a given city.\r\nget_weather = McpToolset(\r\n    connection_params=StreamableHTTPConnectionParams(\r\n        url=&amp;quot;https://mapstools.googleapis.com/mcp&amp;quot;,\r\n        headers={&amp;quot;X-Goog-Api-Key&amp;quot;: os.environ[&amp;quot;MAPS_API_KEY&amp;quot;] }\r\n    ),\r\n)&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4dcc390&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;While the Google Maps Grounding Lite is a Google-managed remote endpoint, the Cityscape agent also demonstrates the other end of the spectrum — a locally hosted MCP server for image generation. The &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;nano_banana&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; toolset connects to the &lt;/span&gt;&lt;a href="http://maps.googleapis.com/mcp" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;GenMedia MCP server&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; using StdioConnectionParams. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With this setup, the agent generates a stylized isometric cityscape image, incorporating the landmarks and weather data gathered earlier. Running a self-hosted MCP server gives you full control over the process lifecycle and environment configuration, but requires &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;a local binary on the host machine or a sidecar container&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;, which adds setup complexity compared to the hosted approach.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;# Self-hosted MCP server: launches the GenMedia MCP server (mcp-gemini-go)\r\n# as a subprocess to generate cityscape images via the Gemini image model.\r\nnano_banana = McpToolset(\r\n    connection_params=StdioConnectionParams(\r\n        server_params=StdioServerParameters(\r\n            command=&amp;quot;mcp-gemini-go&amp;quot;,\r\n            env=dict(os.environ, PROJECT_ID=os.environ[&amp;quot;GOOGLE_CLOUD_PROJECT&amp;quot;]),\r\n        ),\r\n        timeout=60,\r\n    ),\r\n)&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4dcd790&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;ADK supports Google-managed, remote, and self-hosted MCP servers. The former gives you production-ready infrastructure with zero operations overhead, while the latter two offer flexibility for custom or experimental tools.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;2. Enterprise-grade security and content guardrails&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Security in the agentic era can not be an afterthought. Here’s how two key security features can be applied to our Cityscape agent.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Granular control of MCP tools via IAM Deny policies&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Google Cloud lets you control MCP tool access using IAM deny policies — the same governance framework you already use for other Google Cloud resources. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Now imagine we extend the Cityscape agent by adding a BigQuery MCP server — perhaps to query a dataset of historical cityscape metadata or population statistics. The BigQuery MCP server exposes both read-only tools like get_dataset_info and list_datasets, as well as write tools like execute_sql that can modify data.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In our use case, the agent should only query BigQuery for information — it should never execute SQL that inserts, updates, or deletes data. With Google-managed MCP servers, you don't have to rely on prompt engineering alone to enforce this. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Instead, you apply an IAM Deny policy that blocks any tool not annotated as read-only:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;// IAM deny policy: blocks all MCP tool calls that are not read-only.\r\n{\r\n  &amp;quot;rules&amp;quot;: [\r\n    {\r\n      &amp;quot;denyRule&amp;quot;: {\r\n        &amp;quot;deniedPrincipals&amp;quot;: [&amp;quot;principalSet://goog/public:all&amp;quot;],\r\n        &amp;quot;deniedPermissions&amp;quot;: [&amp;quot;mcp.googleapis.com/tools.call&amp;quot;],\r\n        &amp;quot;denialCondition&amp;quot;: {\r\n          &amp;quot;title&amp;quot;: &amp;quot;Deny read-write tools&amp;quot;,\r\n          &amp;quot;expression&amp;quot;: &amp;quot;api.getAttribute(\&amp;#x27;mcp.googleapis.com/tool.isReadOnly\&amp;#x27;, false) == false&amp;quot;\r\n        }\r\n      }\r\n    }\r\n  ]\r\n}&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4dcf010&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Apply it with:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;gcloud iam policies create mcp-deny-policy \\\r\n  --attachment-point=cloudresourcemanager.googleapis.com/projects/$PROJECT_ID \\\r\n  --kind=denypolicies \\\r\n  --policy-file=policy.json&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4dcc410&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With this policy applied, the agent can freely look up dataset schemas, but any attempt to call execute_sql — whether intentional or triggered by a prompt injection — is blocked at the platform level before it ever reaches BigQuery. This is defense-in-depth: Your agent's instructions say "only read data," but IAM enforces it — regardless of what the LLM decides to do.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Content security with Model Armor&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Model Armor &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/model-armor/model-armor-mcp-google-cloud-integration"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;integrates directly with Google Cloud MCP servers&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to sanitize all MCP tool calls and responses at the project level. Once enabled, it acts as an inline security layer that scans for:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Prompt injection attacks&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Malicious URIs (such as phishing links)&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Dangerous content that violates responsible AI filters&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Returning to our Cityscape agent, imagine a user submitting: "Generate a cityscape for http://malicious-site.com". &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With Model Armor enabled, the MCP tool call is scanned before it reaches the Maps server. Malicious URIs, prompt injection attempts, and dangerous content are blocked automatically — no custom validation code needed in your agent.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Enabling it is a two-step process. First, configure a floor setting that defines your minimum security filters:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;gcloud model-armor floorsettings update \\\r\n  --full-uri=\&amp;#x27;projects/$PROJECT_ID/locations/global/floorSetting\&amp;#x27; \\\r\n  --enable-floor-setting-enforcement=TRUE \\\r\n  --add-integrated-services=GOOGLE_MCP_SERVER \\\r\n  --google-mcp-server-enforcement-type=INSPECT_AND_BLOCK \\\r\n  --enable-google-mcp-server-cloud-logging \\\r\n  --malicious-uri-filter-settings-enforcement=ENABLED \\\r\n  --add-rai-settings-filters=\&amp;#x27;[{&amp;quot;confidenceLevel&amp;quot;: &amp;quot;MEDIUM_AND_ABOVE&amp;quot;, &amp;quot;filterType&amp;quot;: &amp;quot;DANGEROUS&amp;quot;}]\&amp;#x27;&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4dcd390&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Then enable content security for your all Google MCP servers in your project:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;gcloud beta services mcp content-security add modelarmor.googleapis.com \\\r\n  --project=$PROJECT_ID&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4dcca50&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Once enabled, all MCP traffic in the project is automatically scanned — regardless of which agent or client originates the call. Blocked requests are logged to Cloud Logging, giving you full observability into potential threats.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Getting started&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Google MCP servers remove the infrastructure hurdles that keep AI agents stuck in prototyping. By combining managed endpoints with platform-level security — IAM deny policies, Model Armor, and Cloud Audit Logs — you get a production-ready foundation with minimum ops overhead. The era of the autonomous agent is here: Make sure your stack is ready.&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;ADK Cityscape agent code repo &lt;/span&gt;&lt;a href="https://github.com/danistrebel/adk-cityscape" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;here&lt;/span&gt;&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Read more about Google MCP servers and supported services &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/mcp/overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;here&lt;/span&gt;&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://codelabs.developers.google.com/ai-mcp-dk-csql#0" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Hands-on codelab&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;: Local to Cloud — Full-stack app migration with Gemini CLI, Cloud Run, and Cloud SQL MCP servers&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li&gt;&lt;span style="vertical-align: baseline;"&gt; Build AI agents with Google &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/run/docs/overview/what-is-cloud-run?_gl=1*i8ohq8*_up*MQ..&amp;amp;gclid=Cj0KCQiA8KTNBhD_ARIsAOvp6DLGEEj0ouZgyTvHN495E7e9huKs2--b0MMYHbttoGeL2-SnKPZkTj8aAqg8EALw_wcB&amp;amp;gclsrc=aw.ds"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Run&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;: a serverless runtime for your agentic AI apps&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;</description><pubDate>Fri, 27 Mar 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/ai-machine-learning/how-to-build-ai-agents-with-google-managed-mcp-servers/</guid><category>Security &amp; Identity</category><category>Application Development</category><category>AI &amp; Machine Learning</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How to build production-ready AI agents with Google-managed MCP servers</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/ai-machine-learning/how-to-build-ai-agents-with-google-managed-mcp-servers/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Lisa Shen</name><title>Product Manager, Google Cloud</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Daniel Strebel</name><title>Solution Lead, Google Cloud</title><department></department><company></company></author></item><item><title>Centralized policy meets distributed logic: Getting to know Eventarc Advanced</title><link>https://cloud.google.com/blog/products/application-modernization/getting-to-know-eventarc-advanced/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Enterprise architects often face a fundamental dilemma: choosing between developer agility and organizational control. Development teams need to move fast and deploy independent microservices without waiting for permission. Security and compliance teams need to be safe, and ensure that data flow is observable and governed by policies.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;That’s why we built &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/eventarc/advanced/docs/overview"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Eventarc Advanced&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;,&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; a serverless eventing platform and the evolution of &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/eventarc/standard/docs/overview"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Eventarc Standard&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. Eventarc Advanced provides&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;an improved architectural pattern for the modern cloud, where &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;centralized policy meets distributed logic&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. By clearly separating the governance layer (the "bus") from the processing layer (the "pipeline"), Eventarc Advanced gives SecOps teams the visibility and control they demand, while freeing developers to choreograph AI agents and build event-driven applications with the autonomy they want. &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/application-modernization/eventarc-advanced-orchestrates-complex-microservices-environments?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Eventarc Advanced became generally available&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in August 2025. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_-_evolution-of-architecture.max-1000x1000.png"
        
          alt="1 - evolution-of-architecture"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In this blog, we take a deeper look at the evolution of integration architectures — from service buses, to microservices, to where we are today — and go into depth with a real-world example. Let’s jump in. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The evolution of integration architectures&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To understand the value of this new pattern, it helps to look at where we came from and why previous architecture patterns forced a compromise.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;The centralized bottleneck of the &lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt;Enterprise Service Bus&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;One early integration architecture approach was the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Enterprise Service Bus (ESB)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, which prioritized centralized control. The ESB emerged to solve the "spaghetti architecture" of point-to-point integrations by providing a centralized communication layer that standardized how disparate systems interact. However, it often introduced serious pitfalls.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The primary issue was what’s referred to as a centralized logic trap. Organizations frequently embedded complex business logic — transformations and orchestration — directly into the governance layer. The resulting middleware layer was opaque, with critical business rules hidden from the developers who owned the services.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Consequently, integration changes typically required the intervention of a central middleware team. Development teams lost autonomy, forced to queue behind integration specialists to ship even minor features, often waiting weeks for updates.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Microservices’ governance gap&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To address this, the industry shifted toward &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;microservices&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; (often described as "smart endpoints and dumb pipes"), distributing logic to give teams the autonomy they were looking for. For synchronous traffic (REST, gRPC), tools like API gateways and service meshes restored a layer of governance by enforcing policies like authentication and rate limiting at the infrastructure level.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;However, as architectures shifted to &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Event-Driven Architecture (EDA)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; for greater resilience and decoupling, a new gap emerged. In a distributed, asynchronous world, centralized control often vanished. This created a &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;governance gap&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; where SecOps teams struggled to maintain order. Three issues emerged to the forefront:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;The visibility void&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Without a central policy, shadow IT services could silently subscribe to sensitive events without detection.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;The policy problem&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Enforcing data residency or PII masking is nearly impossible when the broker treats every message as an opaque blob.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;The dependency risk&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Without clear contracts, changing an event schema risks silently breaking unknown downstream consumers.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;A new pattern: Centralized policy, distributed logic&lt;/span&gt;&lt;/h3&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/2_-_bus-vs-pipeline.max-1000x1000.jpg"
        
          alt="2 - bus-vs-pipeline"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Eventarc Advanced addresses the trade-off between control and speed with a novel architectural pattern: &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;centralized policy meets distributed logic&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Eventarc Advanced maps these distinct responsibilities to two specific architectural resources that each correspond to a distinct role:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;The&lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt; bus:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; This governance layer is a managed, centralized hub where platform administrators enforce global constraints before events are routed. It synthesizes the centralized routing of the legacy ESB with the modern security architecture of a service mesh. It handles Identity and Access Management (IAM), including content-based access control, to strictly define who can publish, and integrates with &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/vpc-service-controls/docs/overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;VPC Service Controls&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to prevent data exfiltration.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;The pipeline:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Think of this distributed, team-owned resource as developers’ integration logic layer. This is where eventing patterns for AI agents and microservices are unlocked, allowing developers to configure event flow and delivery according to their specific business logic. Unlike many service meshes that treat data as opaque bits, the pipeline understands content. Developers can transform events, convert payloads between formats (like JSON to Avro), and configure retry policies and authentication independently.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In other words, by decoupling these duties, Eventarc Advanced provides the &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;control&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; of an ESB with the &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;agility&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; of microservices and the &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;resilience&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; of modern event-driven architectures.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;How it works: A retail event mesh example&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;A typical Eventarc Advanced solution can be implemented with minimal configuration, providing a streamlined experience for both administrative governance and distributed integration logic. To see this model in practice, let's look at a real-world implementation of &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;a &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;retail event mesh&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;.&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Imagine an ecosystem at a global retailer with four autonomous teams in charge of the following services:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Commerce&lt;/strong&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Finance&lt;/strong&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Logistics&lt;/strong&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Intelligence (AI Insights Agent)&lt;/strong&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In a traditional setup, aligning these teams is difficult. The Intelligence team wants access to everything for their models, Finance wants to lock everything down for compliance, Logistics just needs a stable schema to ship boxes, and Commerce needs to roll out new features at a moment’s notice.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;The foundation: Built on CloudEvents&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Eventarc Advanced uses a data model based on the open &lt;/span&gt;&lt;a href="https://cloudevents.io/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;CloudEvents standard&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, which can carry any type of payload. This helps ensure governance and discoverability while retaining flexibility. In our example, before a single event is published, the platform administrator mandates that every message must contain standard attributes and a specific custom extension for governance. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In this example, every event on the bus must carry the following attributes:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;code style="vertical-align: baseline;"&gt;type&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;: Standard identifiers for the event instance (e.g., &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;com.retail.order.created&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;)&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;code style="vertical-align: baseline;"&gt;source&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;: A standard attribute identifying the producer (e.g., &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;//commerce/frontend&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;)&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;code style="vertical-align: baseline;"&gt;data_sensitivity&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;: A custom extension attribute to categorize risk&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In addition, the organization defines three data sensitivity levels:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;code style="vertical-align: baseline;"&gt;restricted&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;(High)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Severe risk data like Credit Card Tokens or Tax IDs&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;code style="vertical-align: baseline;"&gt;confidential&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;(Medium)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: PII like home addresses&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;code style="vertical-align: baseline;"&gt;general&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;(Low)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Safe operational data like Order IDs&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This standardized metadata layer allows the bus to enforce policies based on specific attribute names — checking &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;who&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; sent the data (&lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;source&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;) and &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;what&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; kind of data it is (&lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;data_sensitivity&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;).&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;The workflow&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With this model, the lifecycle of a single order becomes a secure flow where sensitivity changes at every step.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/3_-_flow-no-bus.max-1000x1000.png"
        
          alt="3 - flow-no-bus"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Order placement&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: The &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Commerce&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; service publishes &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;order.created&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; to the Bus. The event’s data sensitivity is tagged as &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;general&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;. The &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;AI Insights Agent&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; service subscribes to analyze market trends.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Payment authorization&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: The &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Commerce&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; service publishes &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;payment.authorized&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; tagged as &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;restricted&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; (containing a secure token). The &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Finance&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; service subscribes to capture the token and executes the charge.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Settlement&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: The &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Finance&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; service publishes &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;payment.success&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; tagged as &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;general&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;, signaling the transaction is safe to fulfill without exposing financial secrets. &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Logistics&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; subscribes to ship the box, and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Intelligence AI Insights Agent&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; is triggered to evaluate market trends for the next supply chain cycle.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Fulfillment&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: The &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Logistics&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; service publishes &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;shipment.ready&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; tagged as &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;confidential &lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;(containing the customer phone number)&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;. The &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Logistics&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; own notification pipeline subscribes to it to trigger an SMS notification.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In a legacy architecture, mixing PCI, PII, and operational data on a single bus would be a compliance nightmare. With Eventarc Advanced, it’s a solved problem.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/4_-_flow-with-bus.max-1000x1000.png"
        
          alt="4 - flow-with-bus"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The bus: the governance layer&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The platform administrator implements a &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;secure strategy &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;on the bus. Rather than blindly trusting internal services, they enforce global policies that inspect these CloudEvents attributes using &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;fine-grained access control (FGAC)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Enforcing source integrity&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To ensure a compromised service cannot spoof events, the bus administrator enforces the producer's identity to match the source attribute.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For example, a bus policy can state that only the principal &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;sa-commerce@retail.com&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; can publish events that match the expression &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;message.source.startsWith("//commerce/")&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;. If the Intelligence AI Insights Agent service tries to publish an event claiming to be from &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;//commerce/payments&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;, the bus rejects the request.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Enforcing a data classification&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To ensure every event is categorized, the bus administrator requires that every payload received by the bus includes a valid sensitivity attribute. A bus policy can check that &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;message.data_sensitivity&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; is one of &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;['general', 'confidential', 'restricted']&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;. This guarantees that the event mesh contains only classified, governance-ready data.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The Pipeline: the logic layer - autonomous team innovation&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With the security posture established on the bus, development teams can then use &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;pipelines&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; to solve complex integration challenges entirely within their own domains. Let’s take a look at a few of these challenges.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Schema-aware formats conversion and payload transformation&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The Logistics team decides to upgrade their warehouse robots to use high-efficiency &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;protocol buffers&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. Instead of forcing the Finance team to change their JSON output (which would break other consumers), Logistics configures a &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;transformation&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; step in their own pipeline.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/5_-_pipeline-json-proto-transform.max-1000x1000.png"
        
          alt="5 - pipeline-json-proto-transform"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;A typical &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;com.retail.payment.success&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; event from Finance arrives as JSON:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;{\r\n  &amp;quot;id&amp;quot;: &amp;quot;89d5663e-789e-4d9f-a65f-f7d83742d987&amp;quot;,\r\n  &amp;quot;source&amp;quot;: &amp;quot;//finance/ledger&amp;quot;,\r\n  &amp;quot;type&amp;quot;: &amp;quot;com.retail.payment.success&amp;quot;,\r\n  &amp;quot;data_sensitivity&amp;quot;: &amp;quot;general&amp;quot;,\r\n  &amp;quot;datacontenttype&amp;quot;: &amp;quot;application/json&amp;quot;,\r\n  &amp;quot;data&amp;quot;: {\r\n    &amp;quot;order_number&amp;quot;: &amp;quot;ORD-2023-8841&amp;quot;,\r\n    &amp;quot;total_amount&amp;quot;: 249.99,\r\n    &amp;quot;currency&amp;quot;: &amp;quot;USD&amp;quot;,\r\n    &amp;quot;transaction_id&amp;quot;: &amp;quot;tx_77382910&amp;quot;,\r\n    &amp;quot;status&amp;quot;: &amp;quot;SETTLED&amp;quot;\r\n  }\r\n}&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4ca5ad0&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The warehouse robots service expects a binary Protobuf message:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;message PaymentConfirmed {\r\n  string order_id = 1;\r\n  double insured_value = 2;\r\n  string currency_code = 3;\r\n  string ledger_reference = 4;\r\n}&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4dcf9d0&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The Logistics team configures their pipeline to accept &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;json&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; as input and output to &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;protobuf&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;. To map the data, they use &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Common Expression Language (CEL)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; to configure a &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;transformation&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;// CEL Transformation to Construct the target Protobuf message\r\n{\r\n  &amp;quot;order_id&amp;quot;: message.data.order_number,\r\n  // 110% of total to cover replacement cost\r\n  &amp;quot;insured_value&amp;quot;: message.data.total_amount * 1.1,\r\n  // Standardize currency to uppercase\r\n  &amp;quot;currency_code&amp;quot;: message.data.currency.upperAscii(),\r\n  &amp;quot;ledger_reference&amp;quot;: message.data.transaction_id,\r\n}&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd4dcdbd0&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This transformation not only maps the input but also applies business logic — calculating the insured value and normalizing the currency code. The Logistics team implements this modernization without a single meeting with the Finance team.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Agentic workflows: Filtering and triggering AI agents &lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Eventarc Advanced enables agentic workflows by allowing pipelines to communicate directly with AI agents using open standard protocols like &lt;/span&gt;&lt;a href="https://github.com/a2aproject/A2A" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent2Agent (A2A)&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://modelcontextprotocol.io/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Model Context Protocol (MCP)&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, while also offering rich capabilities like filtering to optimize when those agents are invoked.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The Intelligence team uses a pipeline they name &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;ai-insights&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; and the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;A2A protocol&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; to connect with an &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;AI Insights Agent&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; that proactively analyzes market trends based on placed orders. Because the agent’s processing is resource-intensive, the team uses a filter to only invoke the agent for high-value orders that warrant deeper analysis.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/6_-_pipeline-filter-mdb-agent.max-1000x1000.png"
        
          alt="6 - pipeline-filter-mdb-agent"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The pipeline filter is configured with the following expression:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;message.type == &amp;quot;order.created&amp;quot; &amp;amp;&amp;amp; \r\ndouble(message.amount) &amp;gt; 5000.0&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd504a690&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;When the filter is passed, the pipeline uses a &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;HTTP Message Destination Binding (MDB)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; expression to directly trigger the agent. By defining a CEL template, the pipeline handles the complexity of constructing a native A2A &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;SendMessage&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; request to the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;AI strategic insights agent&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;. This allows the agent to receive a &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;conversational prompt&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; derived from technical event data without any manual "glue code":&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;{\r\n  &amp;quot;headers&amp;quot;: headers.merge({ &amp;quot;Content-Type&amp;quot;: &amp;quot;application/json&amp;quot;, &amp;quot;A2A-Version&amp;quot;: &amp;quot;1.0&amp;quot; }),\r\n  &amp;quot;body&amp;quot;: {\r\n    &amp;quot;jsonrpc&amp;quot;: &amp;quot;2.0&amp;quot;,\r\n    &amp;quot;id&amp;quot;: message.id,\r\n    &amp;quot;method&amp;quot;: &amp;quot;message/send&amp;quot;,\r\n    &amp;quot;params&amp;quot;: {\r\n      &amp;quot;message&amp;quot;: {\r\n        &amp;quot;messageId&amp;quot;: message.id,\r\n        &amp;quot;role&amp;quot;: &amp;quot;ROLE_USER&amp;quot;,\r\n        &amp;quot;parts&amp;quot;: [\r\n          { \r\n            &amp;quot;text&amp;quot;: &amp;quot;Analyze Order &amp;quot; + message.data.order_number + &amp;quot; for market trends.&amp;quot; \r\n          }\r\n        ]\r\n      }\r\n    }\r\n  }\r\n}&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd7eb0c10&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;A similar prompt message can be crafted for other popular agentic communication protocols like MCP.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This combination of filtering and agentic protocol translation ensures that AI resources are used precisely where they add value. The Intelligence team implements this independently – without writing ingestion code or coordinating with the Commerce or Admin team. The agent can then publish its own strategic recommendation back to the bus, enabling a choreography of AI experts that turns standard cloud events into competitive intelligence.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Advanced API request modeling&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;When a shipment is ready, the Logistics team uses a pipeline to send an SMS using a legacy gateway API. Integrating with legacy third-party APIs often requires writing "glue code" services just to format requests.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The Logistics team eliminates this maintenance burden by configuring a dedicated pipeline to fully construct the exact request expected by the legacy service. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/7_-_mdb.max-1000x1000.png"
        
          alt="7 - mdb"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;They use a &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;HTTP Message Destination Binding &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;CEL expression, which standardizes the phone number and maps it to the &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;X-SMS-To&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; HTTP header required by the API. It also construct the SMS text:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;{\r\n    &amp;quot;headers&amp;quot;: { &amp;quot;X-SMS-To&amp;quot;, \r\n        message.data.phone.matches(\&amp;#x27;^\\\\+1\&amp;#x27;) ?\r\n            message.data.phone : \r\n            \&amp;#x27;+1\&amp;#x27; + message.data.phone \r\n    },\r\n\r\n    &amp;quot;body&amp;quot;: {\r\n        &amp;quot;sms_text&amp;quot;: &amp;quot;Order &amp;quot; + message.data.order_id + &amp;quot; shipped!&amp;quot;\r\n    }\r\n}&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7f5cd7eb3e90&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Finally, they configure a robust retry policy (linear backoff, max five attempts) directly on the pipeline, so that temporary network interruptions don't result in lost notifications. In addition to HTTP endpoints, the pipeline supports guaranteed delivery and out-of-the-box authentication for destinations like Cloud Run, Pub/Sub, Bus, Workflows, and over 200 Google services.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The future of agile integration&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Eventarc Advanced closes an important gap in event-driven architectures: It brings the same level of maturity to asynchronous communication by introducing the pattern of &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;centralized policy, distributed logic&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;For the Platform team&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, Eventarc Advanced provides assurance that a bus can strictly enforce integrity and confidentiality on every message, bringing "service-mesh-like" security to the event layer.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;For the developer&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, it restores autonomy. The pipeline allows teams to filter, transform, convert, and route events to fit their specific needs, enabling them to treat events as first-class products rather than opaque artifacts.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This architecture lays the foundation for the next generation of intelligent applications. A secure, typed, and trustworthy event mesh can serve as the backbone for generative AI agents and real-time analytics, allowing you to safely expose business context to the systems that need it most.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Get started&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Don't let governance slow down your innovation. Here are some Eventarc Advanced resources to get you on your way:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Learn more:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Dive into the full capabilities of the Bus and Pipeline in the &lt;/span&gt;&lt;a href="https://cloud.google.com/eventarc/docs"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Eventarc Advanced documentation&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Get hands-on:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Deploy the "Retail Event Mesh" scenario yourself and explore enterprise patterns with our &lt;/span&gt;&lt;a href="https://cloud.google.com/eventarc/docs/quickstarts"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Quickstarts and Tutorials&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Start building:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Go to the &lt;/span&gt;&lt;a href="https://console.cloud.google.com/eventarc"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud console&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to configure your first bus and pipeline today.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Let's talk:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Have a complex enterprise use case? &lt;/span&gt;&lt;a href="https://cloud.google.com/contact"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Contact Google Cloud Sales&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to discuss how Eventarc Advanced fits into your broader integration strategy.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;</description><pubDate>Fri, 27 Feb 2026 17:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/application-modernization/getting-to-know-eventarc-advanced/</guid><category>Application Development</category><category>Application Modernization</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/0_zjIbf2O.max-600x600.jpg" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Centralized policy meets distributed logic: Getting to know Eventarc Advanced</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/0_zjIbf2O.max-600x600.jpg</image><site_name>Google</site_name><url>https://cloud.google.com/blog/products/application-modernization/getting-to-know-eventarc-advanced/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Milen Kovachev</name><title>Staff Software Engineer</title><department></department><company></company></author></item><item><title>Powering the next generation of agents with Google Cloud databases</title><link>https://cloud.google.com/blog/products/databases/managed-mcp-servers-for-google-cloud-databases/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For developers building AI applications, including custom agents and chatbots, the open-source &lt;/span&gt;&lt;a href="https://www.anthropic.com/news/model-context-protocol" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Model Context Protocol (MCP)&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; standard enables your innovations to access data and tools consistently and securely. At the end of 2025, &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/announcing-official-mcp-support-for-google-services"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;we introduced managed and remote MCP support&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; for services like Google Maps and &lt;/span&gt;&lt;a href="https://cloud.google.com/bigquery"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;BigQuery&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, establishing a standard method for AI to connect with tools, and effectively creating a universal interface for applications. Today, we are expanding this offering to include PostgreSQL with &lt;/span&gt;&lt;a href="https://cloud.google.com/products/alloydb"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;AlloyDB&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;a href="https://cloud.google.com/spanner"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Spanner&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://cloud.google.com/sql"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud SQL&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, as well as  &lt;/span&gt;&lt;a href="https://cloud.google.com/products/firestore"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Firestore&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://cloud.google.com/bigtable"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Bigtable&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; for high-performance NoSQL workloads, and introducing a new &lt;/span&gt;&lt;a href="https://developers.googleblog.com/introducing-the-developer-knowledge-api-and-mcp-server/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Developer Knowledge MCP server&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, which presents an API to connect IDEs to Google’s documentation. These servers run in Google Cloud, providing a secure interface for Gemini and other MCP-compliant clients to easily interact with data and infrastructure.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With the launch of Gemini 3, developers gained advanced reasoning capabilities to plan, build, and solve complex problems. But for an AI model to function as a useful "agent," it must reliably interact with its environment. Today’s announcement extends these capabilities more broadly to the database tools our customers leverage daily as the backbone of their work environment.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To connect your agents to these servers, you don’t need to deploy infrastructure. Just configure the MCP server endpoint in the agent configuration and immediately gain access to your operational data, backed by enterprise-grade auditing, observability and governance. With no infrastructure management, you can scale your agentic workloads without incurring operational overhead.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Bringing operational data to agents&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;These new managed servers enable agents to access specific capabilities across our portfolio:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;AlloyDB for PostgreSQL:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Agents can interact with PostgreSQL workloads, enabling tasks such as schema creation, diagnosing  complex queries for slowness and performing vector similarity search.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Spanner:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; With unified multi-model capabilities in Spanner such as Spanner Graph, agents can model and query complex relationships directly alongside relational and semantic data using standard (SQL and GQL) queries. This allows agents to quickly uncover deep insights (like identifying fraud rings or generating product recommendations) using the MCP tools at its disposal.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Cloud SQL for PostgreSQL, MySQL and SQL Server&lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt;:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Developers and database administrators can use the Cloud SQL MCP Server across MySQL, PostgreSQL, and SQL Server fleets for natural language interactions with the database, AI-assisted app development, query performance optimization and database troubleshooting via agents.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Bigtable:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Bigtable’s flexible schema and high-throughput ingestion capabilities are commonly used for building digital integration hubs and managing time series data. MCP simplifies automating operational workflows and developing agentic customer support, CRM, human resources, IT operations, supply chain and logistics applications with this data.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Firestore:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Focused on mobile and web development, the Firestore MCP server enables agents to sync with live document collections. This supports dynamic interactions such as checking user session states or verifying order statuses via natural language prompts.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Managing applications and infrastructure&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Beyond data retrieval, we are enabling agents to help build and manage applications. The &lt;/span&gt;&lt;a href="https://developers.google.com/knowledge/mcp" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Developer Knowledge MCP server&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; connects IDEs to Google’s documentation, allowing agents to answer technical questions and troubleshoot code with relevant context.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Security and governance&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Connecting an agent to a database requires robust security and governance. These servers are built on Google Cloud's standard identity and observability frameworks:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Identity-first security:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Authentication is handled entirely through Identity and Access Management (IAM) rather than shared keys. This ensures agents can only access the specific tables or views explicitly authorized by the user.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Full observability:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; To track agent activity, every query and action taken via these MCP servers is logged in Cloud Audit Logs. This provides security teams with a record of every database interaction, maintaining visibility alongside ease of access.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Demo: From local code to managed data&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Let’s see these new MCP servers in action.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Imagine an agent designed to automate the migration of a full-stack event management platform for fitness communities. Through a series of natural language instructions in the Gemini CLI, the agent utilizes the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/sql/docs/mysql/use-cloudsql-mcp"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud SQL remote MCP server&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to provision a managed PostgreSQL instance, apply the correct schema, and securely migrate your local data. You don't need to master complex &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;gcloud&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; commands or become a Cloud SQL expert; the agent handles the heavy lifting. This transition is architected in real-time by the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Developer Knowledge MCP server&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, which references official documentation to guide the agent through best practices — easily upgrading your application's backbone from local storage to a fully managed enterprise database.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/1_onemcplaunchblogdemo.gif"
        
          alt="1 onemcplaunchblogdemo"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Support for third-party agents&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Because these servers follow the open MCP standard, they also work with your favorite AI agents. You can easily connect clients like Anthropic’s Claude by adding a Custom Connector in the settings. Simply point it to your Google Cloud database MCP endpoint, and you are ready to start building — no complex configuration files required.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/2_onemcp_launch_claudegif.gif"
        
          alt="2 onemcp launch claudegif"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;What’s next&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We’ll continue to expand this ecosystem in the coming months with managed MCP support for Looker, Database Migration Service (DMS), BigQuery Migration Service, Memorystore, Database Center, Pub/Sub, Kafka and more.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To start building secure, data-driven agents, explore our guides for &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/alloydb/docs/ai/use-alloydb-mcp"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;AlloyDB&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/spanner/docs/use-spanner-mcp"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Spanner&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/sql/docs/postgres/use-cloudsql-mcp"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud SQL&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/bigtable/docs/use-bigtable-mcp"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Bigtable&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, and &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/firestore/native/docs/use-firestore-mcp"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Firestore&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. You can also check out these codelabs for &lt;/span&gt;&lt;a href="https://codelabs.developers.google.com/ai-mcp-dk-csql#0" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud SQL&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://codelabs.developers.google.com/spanner-mcp-server" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Spanner&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, along with this &lt;/span&gt;&lt;a href="https://www.youtube.com/watch?v=SeuhYVg8-AU" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;demo video&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; walking through the app migration to Google Cloud.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=SeuhYVg8-AU"
      data-glue-modal-trigger="uni-modal-SeuhYVg8-AU-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        &lt;img src="//img.youtube.com/vi/SeuhYVg8-AU/maxresdefault.jpg"
             alt="Gemini CLI + Google MCPs: Migrate &amp;amp; deploy full stack apps"/&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-SeuhYVg8-AU-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="SeuhYVg8-AU"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=SeuhYVg8-AU"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;</description><pubDate>Wed, 18 Feb 2026 18:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/databases/managed-mcp-servers-for-google-cloud-databases/</guid><category>AI &amp; Machine Learning</category><category>Application Development</category><category>Cloud SQL</category><category>Spanner</category><category>Databases</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/Gemini_Generated_Image_jcq8tgjcq8tgjcq8.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Powering the next generation of agents with Google Cloud databases</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/Gemini_Generated_Image_jcq8tgjcq8tgjcq8.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/products/databases/managed-mcp-servers-for-google-cloud-databases/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Amit Ganesh</name><title>VP Engineering, Databases</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Rahul Deshmukh</name><title>Senior Product Manager</title><department></department><company></company></author></item></channel></rss>