There's a version of the India AI conversation that happens on stage. At conferences, on panels, in press interviews. It's optimistic, it's big-picture, and it's almost entirely useless as a guide to what's actually going on inside organisations trying to ship AI right now.
The more useful conversation happens when the cameras are off. A group of CTOs at dinner. A founder venting to a peer about a deployment that went wrong. An ML lead trying to explain to a skeptical board why the AI roadmap needs to change again.
That's the conversation worth documenting. Here's what India's AI leaders are actually thinking about in 2026.
Where Does AI Live in the Org?
For a lot of India's larger companies deploying AI, the most pressing question right now isn't technical. It's organisational: where does the AI function actually sit?
Centre of excellence serving all business units? Embedded engineers in each product team? Some hybrid? Every answer comes with tradeoffs that become very visible very fast once you're in production.
Centralising gives you better standards and avoids duplicated work. It also creates a bottleneck and tends to produce solutions disconnected from the actual business context. Embedding in product teams gets you speed and tight feedback loops. It fragments capability and makes it nearly impossible to maintain consistent practices across the org.
The companies furthest along have stopped treating this as a permanent org design question and started treating it as a stage-dependent one. Early: centralise to build standards. Growth: embed to move fast. At scale: federated with strong central governance. Most are stuck at stage one trying to jump straight to stage three.
The Technical Credibility Gap
Here's a tension that shows up in almost every Indian organisation above a certain size trying to move fast on AI. The people who understand the technology well enough to make good calls don't have the authority to make them. The people with the authority don't understand the technology well enough.
This gap existed with cloud, with mobile, with data. It's not new. But AI is moving faster than those waves did, and the decisions being made right now, on infrastructure, on data strategy, on what to automate and what not to, are harder to reverse than they were then.
The leaders handling this well aren't trying to turn every executive into a technical expert. They're building a shared language and shared mental models so that technical and non-technical people can reason from the same starting point.
- Regular "AI without jargon" sessions with senior leadership, actual conceptual conversations, not product demos
- Technical leads who can move between engineering reality and business context without losing either
- Decision frameworks that separate what the technology can do from what you should actually do with it
GTM for AI-Native Products Has Changed
For India's AI-native startups, the ones building products where AI is the core value prop rather than a feature, the GTM conversation in 2026 has shifted. A year ago it was about explaining what AI could do. Now it's about proving ROI before the runway runs out.
The early adopter wave is over. The enterprise buyers who got excited about AI for its own sake have already bought something. The next cohort is more skeptical, more ROI-focused, and most of them have watched at least one AI initiative underdeliver at their company or a peer company.
"The enterprise buyer has been burned once. You're not selling AI anymore. You're selling proof that this particular AI actually works."
That changes the motion in a few specific ways:
- POCs need to be shorter, tighter, and tied to metrics the buyer already reports on internally
- Your internal champion needs to be able to explain the product to their CFO, not just their CTO
- Integration matters as much as capability. AI that doesn't fit the existing workflow just doesn't get used.
- Success metrics have to be agreed upfront, before the POC starts, not retrofitted to explain results afterward
The Talent Picture Is Getting Complicated
A year ago, the talent conversation in Indian AI was almost entirely about scarcity. Not enough ML engineers. The good ones were getting poached at aggressive multiples. Hiring was the bottleneck on everything.
That's still partly true but the picture has changed. Deep ML talent is still hard to find. But the definition of what counts as AI talent has expanded, and with it the hiring math has gotten messier.
- How much ML knowledge does a product engineer actually need to be effective on an AI team? The bar is higher than it was a year ago.
- What do you do with strong engineers who are resistant to AI-augmented workflows? This is a real management problem and it's growing.
- How do you actually assess AI fluency in candidates when every resume now claims it?
- What's the right mix of ML specialists and AI-fluent generalists on a product team?
Nobody Knows How to Measure This Yet
The most candid conversation in Indian AI leadership circles right now is about measurement. How do you actually know if your AI initiative is working?
It sounds like it should have an obvious answer. It doesn't. The easy-to-measure metrics (cost savings, time reduction, throughput) often miss the value that matters most. The judgment calls AI is enabling. The decisions getting made faster. The mistakes that aren't happening anymore. These things are real but they're hard to pin down cleanly.
At the same time, the metrics that actually capture the right outcomes are slow to accumulate, hard to attribute specifically to AI, and easy for skeptics to pick apart.
Teams that only measure what's easy end up optimising for the wrong things. Teams that wait for perfect metrics end up with nothing. The ones doing this well have started with imperfect metrics and committed to making the measurement system better as they learn what they're actually trying to move.
The India Opportunity Is Bigger Than the Headlines Suggest
Under all these tactical questions sits a bigger strategic one that India's best AI leaders are starting to take seriously: the Indian market is not a smaller, cheaper version of the US market. It has its own structural characteristics that create AI opportunities at a scale that doesn't exist anywhere else.
India's economic formalisation is still happening in real time. Hundreds of millions of people and tens of millions of businesses are entering digital financial and commercial systems for the first time. AI built natively for this transition, for users who are voice-first, low-bandwidth, switching between eight languages, formalising their first business, is building something that doesn't exist in any other market.
The people who understand both the technology and this market deeply are sitting on a real opportunity. The conversations that help them figure out what to do with it are worth having.
That's what the Cabal is here for.