Krishna K.

EdTech 1.0 never had a learner problem. It had a CAC problem. Education DPI is about to change that math. Here's how I see it. India's education DPI standardizes three things: → Student identity (APAAR ID) → Course discovery (ONEST, built on the Beckn protocol) → Verifiable credentials (DigiLocker, Academic Bank of Credits) Onboarding, discovery and credentialing become low-cost public utilities. That is not free demand. You still have to earn the learner. But the plumbing is no longer your cost. Christensen called this the Law of Conservation of Attractive Profits. In his words: "when modularity and commoditization cause attractive profits to disappear at one stage in the value chain, the opportunity to earn attractive profits with proprietary products will usually emerge at an adjacent stage." DPI commoditizes the base layer. So the margin moves up. The filter Let's be precise. DPI does not make switching costs zero. It makes data lock-in zero. When a learner's ID, records and credentials are portable, nobody stays because their certificate is stuck inside your app. ✕ Thin wrappers that host content or hold certificates hostage lose their only moat. Mediocre tools will churn. ✓ Products that own a real workflow keep their switching costs. Earned, not trapped. And they inherit near-zero onboarding cost. You can't outspend this. You have to out-deliver. Where the margin moves 1. Institutional middleware Most colleges and schools don't have the engineers to integrate with APAAR, ABC and state registries. Middleware that connects legacy campus ERPs to the DPI stack.. credit transfer, compliance, reporting.. is a real business. 2. Instant skill verification Degree checks are still slow and manual. An API that verifies credentials from inside the employer's ATS in seconds replaces weeks of back and forth. 3. Consent-based learner financing Adult learners in skilling programs often have no credit history. With explicit consent, verified learning records can become a credit signal. Adults only. Consent first, the Account Aggregator way. No quiet monitoring of children's data. DPDP is clear on that, and it should be. 4. AI that plugs into the rails IMHO the biggest one. Tutoring and assessment that reads a consented, portable learner record, adapts to the student, and writes verified progress back. The moat is learning outcomes, not stored data. The takeaway If your moat is trapped data, DPI will take it away. If your moat is workflow depth and measurable ROI, DPI just cut your onboarding cost to near zero. Which of these four gets built first in India? Curious what builders and investors here think. #EdTech #DigitalPublicInfrastructure #B2BSaaS

11

2 Comments

Working on AI agents lately, one realisation I have is prompts ≠ governance. You can tell a model “Don’t do this” but once it starts calling APIs control gets blurry fast. Really like this Microsoft’s Agent Governance Toolkit. It finally adds a real “stop” where it matters. Feels like we’re moving from “hey it works” to “we trust it.” https://lnkd.in/g9xXS4XN

36

This caught my eye because Malayalam, my mother tongue, is used as one of the examples to illustrate the “tokenizer tax.” Great look at how the way AI models tokenize different languages can affect both cost and capability. Language representation in AI goes well beyond simply being able to understand a language. https://lnkd.in/gMzzTxSd

21

Want to run effective meetings? Add Facilitator to your Teams meetings. It keeps your agenda on track, captures live notes as you talk, manages Q&A so nothing gets lost, and generates an AI summary afterward. No more meetings that drag on or end without clear next steps. Check it out: https://lnkd.in/gXyesa-W #MicrosoftTeams #TeamsAI #ProductivityTools

63

For the past several months, my team has been working in a slightly different way — and only recently did I pause to reflect on how meaningful that shift actually is. AI tools are no longer just experimental or reserved for pilots and hackathons; they are becoming a fundamental part of our daily processes. This transition to treating AI as a baseline technology has transformed the nature of our work. My team has shifted left in utilizing AI—integrating it earlier in design, development, and validation. This has led to a corresponding increase in velocity, with speed and confidence becoming more apparent downstream. In a data‑centric world, success depends on clarity at every layer — clean schemas, predictable pipelines, and minimal transformation friction. When data workflows are brittle or ambiguous, AI doesn’t fail loudly; it delivers diluted or misleading results. Our progress in this phase has not stemmed from pursuing every new tool but from rethinking how our workflows operate from start to finish, with discipline, intent, and a systems mindset. The key takeaway for me is that AI does not replace engineering rigor; instead, it elevates the standards we must uphold. #AINativeEngineering #AIEngineering #GenAI #AIWorkflows #EngineeringLeadership

24

1 Comment