Ameya P on how Indian IT can flip the AI script

Between existence, survival, and the possibility of adaptation.

Hi folks, Pranav here.

You (and maybe your portfolio too) are probably tired of hearing this, but the age of AI agents has been a thorn on Indian IT’s side. Its business model is getting stale each passing day. But what we often don’t understand is what this means for the day-to-day nitty-gritties of the industry.

Now, there is a possibility that Indian IT might just adapt to the new paradigm. We’ve briefly covered some semblance of this possibility in past Daily Brief stories. And usually, we can’t say for sure what such a possibility even entails. After all, what does adaptation for Indian IT look like? What are the forms of inertia they will have to overcome to successfully change themselves? And even if they do adapt, will they be able to defend their business from the new threats that AI will inevitably give birth to?

To answer these questions, we had a nuanced conversation with Ameya Pimpalgaonkar, a storied professional with two decades of experience in the global IT industry, working across giants like IBM, Accenture and Infosys. He has donned many hats, from software developer to co-founder and CTO. He’s also a prolific technology investor in both public and private markets, and is very well-known on X for his investing takes. We’ve quoted his work in the past on The Daily Brief. I highly recommend reading his newsletter.

Personally, I learnt a lot. Most mainstream conversations about Indian IT are either a black-box, or are too one-size-fits all. Ameya’s answers cut through both of those problems succinctly. The idea was not to stick to a highly doomer or highly optimistic narrative, but weave a story that pinpoints the opportunities, the threats, and understanding the depth of how much AI can commoditize any possible future moat.

You can listen to this conversation on YouTube, Spotify, or Apple Podcasts. The transcript of the podcast is below.

Please note: This transcript has been generated using AI, and may contain minor errors or inaccuracies.

---

Pranav Manie: My first question is around what we’re seeing in recent earnings calls — you’re finally starting to see a lot of the signals that people said would indicate AI adoption on some level.

Pranav Manie: The move from headcount-based billing to outcome-based pricing — even though that’s not happening as quickly as anybody would imagine — is still happening, and a lot of new revenue lines seem to be opening up that Indian IT firms are jumping into. But even if all of that happens, what is the moat that any of these companies, especially the biggest four, actually have? What defensibility do they have in any of these new revenue lines that AI is opening up?

Ameya P: The shift from headcount-based to outcome-based pricing that you’re describing — the shift itself is very subtle. That discussion is coming out in the boardrooms, but it’s not that clear when it comes to actual employee engagements.

Ameya P: Until recently, if you look at AI, it was primarily a cost story. Like for example: we are going to use AI to improve our margin. That used to be the common theme across boardroom discussions. But in my view, what has changed in the last two to four quarters is that you’re now hearing a revenue angle coming into that discussion. Top tier companies have all started to book something called AI-led transformation deals as a separate revenue category.

Ameya P: If you look at the cloud paradigm, especially after COVID, a lot of companies recategorized their ongoing deals with customers into digital revenue. How did they do the categorization? Nobody knows on what basis they did that.

Ameya P: Nobody knows, right? Some client moves from on-prem to cloud, nothing actually changes — and you start calling that cloud revenue. That’s not a proper representation of cloud revenue. The cloud transformation that actually brings value is what brings more revenue and leads to expansion of wallet share.

Ameya P: At a given account, that transformation looks very different. You don’t just go from on-prem to cloud and start calling it a different revenue stream. That’s the very typical narrative that so many companies use. But what we have to focus on is the structural signals, not the marketing ones.

Ameya P: Most of these, in my view, are still time and material engagements with an AI app around it. The model hasn’t changed much. What’s changed is what’s labeled on the invoice. But the real shift — the outcome-based pricing — is very interesting because customers are now asking questions very differently.

Ameya P: Earlier questions used to be: I want to do this, how long is it going to take, what is your T&M strategy, what is the duration? There used to be benchmarks. If you’re in a multi-country rollout, your first payment could come when the pilot goes live and the hypercare around it is finished.

Ameya P: Your second payment comes when the next 10 countries go live, and the final payment when the last set of countries go live. So the benchmark-based payment was always there — even in a T&M model. The shift now is: okay, we will go live with you, no doubt, but what can you do differently to reduce the cost of maintaining systems after go-live? I want to operate with a very lean org structure, a very lean vendor structure.

Ameya P: What can you do differently at that point? This is exactly where outcome-based pricing is coming out — because it requires the vendor, meaning the IT companies, to absorb the delivery risk. That’s the biggest point. Until now, the delivery risk was always covered because it was time and material.

Ameya P: Even if there was a delay, there was always the possibility to raise a change request and cover what you missed. That’s not the situation anymore with outcome-based. You bear the risk on your own books. That’s a significant shift in how these companies operate.

Ameya P: Indian IT’s entire operational model is built on de-risking delivery through headcount elasticity. If you’re not able to deliver with two people, you add five more. That was the way to de-risk delivery — but that is now changing.

Ameya P: You add people when the scope grows. You can’t do that when you’ve sold a fixed outcome at a fixed price.

Pranav Manie: Understood. Yeah.

Ameya P: Now, the transition from headcount to outcome is, in my view, going to be slower than how it appears. It’s being discussed very actively on social media and in client discussions, but the actual pace of change is going to be slow for the next year or so — and there are reasons for that.

Ameya P: First, the contractual challenges. All service level agreements are still tied to response time, not to a business outcome. If you go to Jira, or any ticketing tool like ServiceNow, you would clearly see the service catalog — the SLAs as per the category of an incident — and the SLA translates into how many hours that incident should be resolved in. When that time elapses, you get the escalation emails automatically. But inherently, this is still tied to response time. It is not tied to the quality of the fix being delivered.

Ameya P: There is no connection to the business outcome defined yet. And what this means is you have to renegotiate a contract that is already a legal document — an already time-tested relationship with the customer. How would you even define a business outcome? That’s the biggest challenge right now.

Ameya P: Second, the internal P&L structures. These companies measure utilization — bill hours per engineer, revenue per headcount, and so on. Outcome-based pricing blows up that matrix entirely. You’d need to rebuild how delivery units are measured, what the new unit of measure is, how to incentivize it, and how to report on it.

Ameya P: Analysts know they’re going to ask these questions — they know outcome-based pricing is coming — but they don’t know how to measure it. In my view, that represents at least a couple of years of internal transformation. And it’s not just a pricing change. The most important piece is client readiness.

Ameya P: In my 20 years of career, what I’ve seen is that whatever scope of work you do in the first year — or even the first six months — with a customer engagement, it always balloons four to five times compared to the initial SOW.

Ameya P: So what you end up delivering is, I think, four to five times more than what was defined in the SOW.

Pranav Manie: And that’s where the delivery risk falls in place as well, correct?

Ameya P: Correct. Exactly. Since most enterprise customers can’t define what success and closure looks like in the early phases of SOW definition — that’s what I mean by client readiness. So the transition is definitely real, but it’ll happen deal by deal. It won’t happen as a wave.

Ameya P: Now, when it comes to the new revenue lines — where exactly does the opportunity lie for these companies? There are two clear ones, and one we can already see — TCS is taking that approach.

Ameya P: They’re going the AI factory, cloud service, or data center route. They’re building their own AI infrastructure plays, and the intent is to become the managed AI ops layer for enterprises — particularly for customers who don’t want a hyperscaler dependency. The margin profiles are different, but those businesses are relatively stickier.

Ameya P: That’s one revenue line opening up, and we already see TCS setting a precedent with their data center approach. The second approach I’d like to talk about — if you look at the nineties, the IT boom, and what happened in the next 15 to 20 years — let’s call that paradigm the legacy software paradigm.

Ameya P: Let’s divide it into two categories. Legacy software is anything pre-LLM. New software is anything post-LLM. In the pre-LLM era, all legacy software had a very specifically defined playbook. There would be a software OEM — Microsoft, SAP, Oracle, anyone like that.

Ameya P: They would control the enterprise software space. They’d sell a license — a perpetual, multimillion dollar license deal. Then, to customize it and make it work for a given customer in a given business domain, with all their custom processes, you would need software integrators.

Ameya P: You would need the Infosyses and TCSs of the world. That was the paradigm — enterprise software OEMs and legacy software integrators. Now translate that exactly into today’s parlance. LLMs are, in a way, taking the place of legacy software OEMs.

Ameya P: Cloud providers are launching new services constantly and getting deeper into enterprise. Many of these LLMs also have associations with legacy software providers to penetrate the enterprise space. So then what is the definition of SI in this new world? People will say that LLMs will eventually be built, but the application layer is still serviced by India.

Ameya P: I agree — that’s probably one of the easiest proxies anybody can point to. But the most important and value-additive proxy that very few are talking about is LLM fine-tuning. The way this transitions is: an LLM comes into the picture trained on a generic set of data, able to generate outputs up to a certain cutoff date.

Ameya P: Anything that comes after that cutoff requires a web search. So anything not in the LLM’s training data, it fetches via a web search — it’s doing a RAG. But the point is, you cannot improve LLM hallucination and fidelity just by doing RAG.

Ameya P: Your costs will balloon. What you need is reinforcement learning of the existing LLM on a very specific domain dataset. To do that, you start with a generic pre-trained LLM and do reinforcement learning to fit it into a certain industry or domain.

Ameya P: Wipro has gotten into this. They already have about 13,000 data OTRs on their payroll. What these people are doing is taking a baseline LLM and creating reinforcement learning workflows for the LLM OEMs. And Wipro is just one example — there are other companies in the private space doing this for STEM as well.

Ameya P: There’s Scale AI at the global level. There’s a company in India called [unclear], and more companies will eventually come and build IP in this space. Data pipeline preparation and reinforcement learning workflows — that’s where the next alpha lies for all these companies.

Ameya P: Why? Because these IT companies have data that has been anonymized and pseudonymized over 20-plus years. It’s properly stored and structured so that no personal information or customer record can be attached to it.

Ameya P: Any data that’s anonymized and pseudonymized is perfectly suited to be transformed. This is where data labeling richness comes in. A surgeon looks at an X-ray and says, okay, there’s a nodule in the chest — that line by the surgeon is an annotation labeling that data.

Ameya P: If you just pass that X-ray to an AI, the AI will say this is this, this is that — but it’s missing the surgeon’s label, the surgeon’s context for interpreting what it means. That data labeling, in my view, is going to be one of the quickest revenue streams opening up for many of these companies.

Ameya P: What it means in practice is: you take models with pre-trained weights, bring in domain experts from legal, geography, STEM, and so on, and have them engage. You must have experienced this — whenever you use ChatGPT and ask a strategic question, sometimes it generates two outputs and asks which one you prefer.

Ameya P: That’s exactly ChatGPT doing reinforcement learning to identify your preferences. Now think about doing that at an LLM level, at an enterprise level. India has the workforce and the muscle to do this. It’s not glamorous, but it’s real revenue.

Pranav Manie: Revenue. Right.

Ameya P: Exactly.

Pranav Manie: Is this also something that Palantir does in the defense industry, so to speak? I might be wrong — I doubt they have that kind of human muscle — but is this reinforced learning on a domain-specific dataset something they do for, say, the Pentagon?

Ameya P: Sorry, can you come again please?

Pranav Manie: This entire approach you’ve described — where you need reinforced learning on a domain-specific set of data. I don’t know whether Palantir has data annotators on their payroll, but I’m assuming this is something they do for the Pentagon.

Ameya P: Yes, of course. Companies protected by defense boundaries can’t go to external vendors — they have to build internal teams. But at the same time, from my interactions with startups operating in the defense space, I see them onboarding a lot of retired army officers, navy senior officials, and using their expertise for data labeling.

Ameya P: That’s already happening in India. And I see another line of businesses coming out of that space. It all starts with something very basic — I will label your data. But eventually, it’s a newer version of BPO.

Ameya P: The real value comes when you build a platform around it in a multilingual, multimodal way. If you do that, you get to a point where every single LLM is reliant on you before releasing a new version. That’s where the sticky value and high margin space exists.

Pranav Manie: So — not that I know what these platforms are actually going to do — I don’t know what Infosys Topaz does, I don’t know what Wipro AI 360 does or what market they’re going after. They’re called agentic AI platforms, but I don’t totally understand whether they’re a collection of agents or the kind of platform you’re describing.

Pranav Manie: What’s your understanding of these attempts?

Ameya P: Topaz, I would say, is more of a branding umbrella than a single product. It’s a mix of multiple products, announced in 2022 or 23, I believe. What it does is wrap together a host of AI services, tools, accelerators, and partnerships under one label — Infosys Topaz. It’s not really a software platform in my view.

Ameya P: It’s more of a go-to-market organizational approach. What’s actually inside it is multiple layers. The first is the service layer — consulting and implementation services for enterprise AI, still largely people-led, just relabeled as an AI-first service layer.

Ameya P: Then comes Cobalt. Topaz’s product architecture sits on top of Cobalt, which is their cloud platform play. The pitch is: cloud modernization plus AI activation, combined. And then there are pre-built AI accelerators — industry-specific templates and workflows for BFSI, retail, manufacturing.

Ameya P: These are meant to reduce your time to deploy for commonly defined AI use cases. They’re real use cases, but not deeply proprietary. And then comes the partnership ecosystem — heavy co-branding with Nvidia, Microsoft, and so on.

Ameya P: So Topaz is partly a vehicle to resell and integrate hyperscaler AI capabilities with Infosys’s wrapper around it. When all IT companies talk about a bundled product, this is exactly how it works — it’s a layered product. It’s not one platform you plug in and start using tomorrow.

Ameya P: It has a service layer, a platform layer, a tools and accelerator layer, and a partnership ecosystem attached to it.

Pranav Manie: Got it. There was a tweet of yours that I came across — which is also where this question comes from. You differentiated between large cap Indian IT with pyramid-shaped organizational structures versus smaller companies with diamond-shaped structures. How much does this org structure matter when it comes to AI adoption?

Ameya P: There are two things here. The pyramid structure is what Indian IT was built on — a lot of junior engineers at the base, very thin middle management, and an even thinner senior layer at the top. It was perfectly designed for labor arbitrage.

Ameya P: The model was simple: hire bulk at the base, bill at a markup, and use senior people to manage client relationships and escalations. The pyramid was designed to drive margin through high utilization at the bottom.

Ameya P: Now with AI, it impacts the base of that pyramid — junior analysts, junior developers, QA testers, reporting analysts. These are exactly the roles people claim will be replaced by AI agents in the first wave. When the base shrinks, the whole margin math goes for a toss.

Ameya P: But the bigger problem — as you correctly said — is that the pyramid structure creates inertia. Every senior person at the top was at the base at one point. The entire career contract inside these firms is: you start at the bottom, grind through, get promoted, and move up the pyramid over 8, 10, 15, 20 years.

Ameya P: That doesn’t happen with AI anymore. But the challenge is — if junior roles disappear, where do future seniors come from?

Pranav Manie: Correct.

Ameya P: We still don’t have a clean answer to that. And if you look beyond the top 4, 5, 6 IT companies to smaller companies that have always run leaner structures, they’re much better positioned to take advantage of this.

Ameya P: One reason is they don’t have the overhead. There’s no risk of firing large numbers of people and creating chaos in the market. Instead, they can just increase the productivity of existing employees — and I know for a fact that’s happening. For smaller companies, it’s much easier to adopt AI tools.

Ameya P: These companies are starting to convert their product features into markdown files. Every product backlog, every PI planning exercise — the output is a markdown file today, which feeds directly into whatever coding copilot you’re using.

Ameya P: And once you have everything defined in a markdown file, building a feature becomes so much easier.

Pranav Manie: Yeah.

Ameya P: So smaller companies are much better prepared to benefit from this. Just getting access to a copilot at a top four or tier-five company is a nightmare.

Ameya P: Coming back to the diamond structure — it assumes a couple of things. First, that junior work is largely going to be automated. Second, that value creation is in judgment, not execution. That’s a very important point.

Ameya P: When value is in judgment, you’re valuing your mid and senior levels more than the execution layer. And that ties back to selling outcomes, not resources. In a diamond-shaped org, your model is outcome-based, not resource-based.

Ameya P: And the biggest challenge is you can’t flip the pyramid overnight. TCS has 600-odd thousand employees, Infosys has 300-odd thousand — you simply can’t re-skill even a hundred thousand junior engineers into mid-level AI solution designers.

Ameya P: And when companies say they have an AI-ready workforce, it’s nothing but an internal training that’s been done. They don’t understand the basics — the nitty-gritty around LLMs, how important it is to provide context. People just start chatting with a copilot in natural language and burn through their entire token quota.

Ameya P: They don’t understand how important it is to build context — not just through prompts, but through other approaches as well. What is a chunking strategy? How does it impact your cost? Eventually it all ties down to cost. When you’re in a customer engagement selling an AI solution and you don’t know what will hit your costs, you’re not going to win that contract. It’s as simple as that.

Ameya P: These inertias are not technological. They’re much more than that.

Pranav Manie: It’s much more than that. Yeah — it rarely is technological, in my view.

Pranav Manie: One of the questions I had came from a podcast on The Ken, where this entrepreneur named Sidu Ponnappa — who also runs an IT services firm in the AI space — was doing pretty well. He had said that all of these companies — like, for example, Infosys signing a deal with Anthropic, or with OpenAI — are most likely software procurement deals where the goal is to roll out cloud tools across the enterprise. And like you said, they’ll all be sandboxed. But the analogy he put forward is that it’s not enough to procure the best model available at any given point in time, because the best model is very likely to change in the next week, or even the next day.

Pranav Manie: The only way to survive is to build an operating system around how you work — to know how flexible you can be. Like you don’t know what task works best for Codex versus Claude Code. So how could you ever claim to have an AI-native strategy?

Pranav Manie: So my question — which I think you’ve already partly answered — is: how much does Infosys signing with Anthropic actually matter as a signal?

Ameya P: It’s very important that these enterprise companies announce these deals — they’ve taken a lot of flak for not doing anything on AI. But in my view, the deals themselves hardly mean anything structurally. Even if you release Claude Code or Codex internally to your employees, these tools are good for building a POC or a pilot. But taking that to production is a very different journey. The compliance and governance side, the legality on the customer side — most customers of Indian IT are from Europe and the US, and they have very strong legal boundaries around AI and what gets shipped into production.

Ameya P: You simply cannot code something with AI, do a QA, and ship it into production. So in my view, all these deals are good as narrative-building. They give employees access so they can start building — because that’s also important.

Ameya P: Unless you build something, you won’t know what to do and what not to do, how to structure your context better, how not to burn tokens. If you’re building with Claude Code, for example, and you chat with the copilot asking to build something, you’ll have a session. After the session ends, Claude Code compresses it so it can refer to it in the next interaction. Then you come back the next day, say “I don’t think this is working,” and the copilot goes through all your files and the entire project structure.

Ameya P: That’s a waste of time. Unless you have access to Claude Code and make those mistakes — figure out that you’re burning a hundred thousand tokens every interaction and understand why — all that learning won’t come. And that’s what comes out of these deals. That’s how I read it. These deals are not intended to ship code to production faster — we’re not there yet. They’re about letting employees learn: how to use Claude Code, how to use Codex, how to code better, how to structure projects, how to version properly.

Ameya P: How do you avoid consuming tokens unnecessarily? How do you keep building context? How do you organize your CLAUDE.md? How do you create markdown files, governance, naming conventions at a project level? All that hands-on, dirty learning that every engineer goes through in their first year — these deals are enabling that. That’s how I’m reading it, not as a path to shipping enterprise products faster.

Pranav Manie: We have to make the mistake first.

Ameya P: Exactly. These companies are willing to do that. And I think that’s a positive thing, rather than calling them out with “what deal?”

Ameya P: This deal is not going to result in top-line growth immediately. Top-line growth doesn’t come like that. You have to re-skill in line with what new customers are demanding. This is all a re-skill initiative — and eventually something good will come out of it.

Pranav Manie: Makes sense. That’s a more optimistic frame than what I’ve heard elsewhere.

Pranav Manie: I had a twofold question around how AI changes go-to-market. Historically, you’ve always had a salesperson or sales team go to a client, work a demo with them — a demo that obviously takes a while to build — show it after months, and then the client signs a contract with the systems integrator. But now there’s a possibility that a salesperson could build a working demo of what you want right there in the call itself. Doesn’t even need to know a lot of code — if the goal is a simple CRM, they can do it. Some companies already seem to be doing this. So what’s your view on how client relationships are changing in the AI space? And how does this differ for large cap companies versus smaller mid-cap ones?

Ameya P: That’s an interesting question. My answer is twofold. Within enterprise clients, I don’t think it’s easy for a sales guy to just go build a demo and show it to the customer. I’ll tell you the reality — within enterprise or any IT company, delivery guys and sales guys have a classic tension, because sales guys always over-promise.

Ameya P: They promise things that are technically not possible to build — or that can’t be built in the time they commit to. I’ve been in that situation — not just in Indian IT, but even in product companies where forward deployment engineers always vent at sales guys for over-promising.

Ameya P: They say our product can do this or that, but in reality there’s no feature at the base layer that can do it. You’d need to raise a patch note to the core engineering team to build the foundation, and then build the feature on top. That’s the reality.

Ameya P: All the respect to sales guys — they bring in revenue and contracts — but there’s a real disconnect between understanding what can actually be delivered in a production environment versus what you need to show in a demo to get a contract.

Ameya P: The delivery will get faster and deeper — I’m not denying that. But expecting a sales guy to code a CRM live in a customer meeting — that’s a bit of a stretch.

Pranav Manie: — you’re not expecting that to happen anytime soon.

Ameya P: Too much of a stretch, at least in the enterprise space. Outside of enterprise — where sales cycles are shorter because the goals and business processes are entirely different — it’s probably possible. But in my view, it’s not a realistic scenario for at least the next year, where a sales guy adjusts a demo in a meeting and walks out with a contract.

Ameya P: Sales cycles are not technology-driven. They’re driven by intangibles — delivery quality, what happens after delivery. And in the age of AI, the importance of audit trail is something we can’t even begin to describe. I was in a call yesterday with a company where they discussed an incident — they had an army of agents, and three of them didn’t agree with what the rest concluded. The system just froze.

Ameya P: The product had no way to determine a resolution. Who has the final say? Who says “okay, this is it”? Even the orchestrator agent couldn’t come out of that state. And it froze even though the product had a built-in rule — if agents disagree, do this, do that. Agents were self-aware, and it still happened.

Ameya P: We’re still discovering new aspects of working with these tools. You need a human coming in to handle the dirty work. But do you have that human if you’ve already let people go?

Ameya P: It sounds flashy — this idea of the solopreneur and all that — and eventually we will get there. But we’re not there yet. In enterprise, what sells is the relationship, the plumbing around the technology. Technology sells itself. What are you going to do with that technology?

Ameya P: How are you going to assist a customer in a dire situation? How are you going to manage something completely unpredictable? How are you going to navigate working with a regulator, compliance requirements, data boundaries? It’s become even more challenging in the world of AI. Imagine a production environment where a screenshot is taken and sent to an agent — what if that information leaks? Who owns that? Who’s accountable if the agent makes a wrong decision?

Ameya P: We don’t have answers to those questions yet. So the sales cycle and GTM won’t massively change around decision-making. Where it will change is in proposal generation — that’s going to be largely automated. The 40-page proposal that used to take a Solution Architect a week? That gets done in a day. You already have pricing tables, boilerplate, case studies — you know what the customer is looking for. You can generate that proposal in three or four hours.

Ameya P: Then send it to a solution architect to evaluate and correct, so you don’t hallucinate through the process. Certain pockets of the GTM process will get expedited. The classic cycle used to be: prospect → qualified lead → discovery → solution design → demo → proposal → legal → close. Of that, the prospect phase and the legal phase are going to be very expedited — because there are a lot of legal startups in the AI space where the LLM drafts the initial document and you pass it to senior advisors.

Ameya P: Who then review it and pass it on.

Pranav Manie: So you’re saying it’ll still be relationship-driven, at least for the foreseeable future — as it always has been.

Ameya P: In fact, trust becomes even more important when you’re selling an agentic AI system. The pitch is: if something goes wrong tomorrow, I’m there to make sure nothing slips. That’s why relationships are even more critical now.

Pranav Manie: And you’re saying this is true for large and small companies equally?

Ameya P: Yes. Everywhere. Accountability becomes the relationship. That’s my understanding.

Pranav Manie: And I’m assuming this would be even truer for sectors like banks, which obviously have heavy-handed regulation.

Ameya P: Absolutely. Especially with banks and fintechs — due to compliance and regulatory constraints, they can’t just switch vendors at will. And it’s very important to understand that you cannot whitewash everything with AI. If something is working well, if something is stable and running without AI, why go wrap it with AI?

Ameya P: You don’t have to. I understand you’re trying to identify process inefficiencies and bottlenecks — but you cannot improve a banking workflow beyond a certain point. The regulatory and compliance bottlenecks exist for a reason: to protect the common person, the infrastructure, and our digital sovereignty.

Ameya P: We can’t afford to have vulnerabilities shipped into production, especially when we don’t fully understand how these systems work. You must have seen what happened at McKinsey recently?

Pranav Manie: Mm. What happened?

Ameya P: There was a cyber attack. The attacker was able to access about 700,000 documents, including millions of chats, chat logs, and systems — all of which required no authentication.

Ameya P: At McKinsey — yes, at McKinsey — there was no authentication in place. And they found places where SQL injection was possible. The time and cost for the attacker? Two hours and $20.

Ameya P: And this is a classic example of why you can’t just go heavy on AI to make noise and support your stock price. Eventually something comes and bites you.

Pranav Manie: Awesome. Great. You touched on digital sovereignty, which is a very nice point to close on — at least in earnings calls, we’ve been seeing deals that reflect where the world is headed. Every country is prioritizing national security of some kind. Middle Eastern companies are building data centers because they have the capital and don’t want hyperscaler dependency.

Pranav Manie: You’ve written on digital sovereignty — how banks and IT companies are thinking about it. Where do you see AI and digital sovereignty colliding? Where do they gel together, if they gel together at all?

Ameya P: They definitely gel — especially in the banking and FinTech space. At the customer experience level, customer onboarding, customer experience — all of that, AI gels fantastically. Where they don’t gel is the layer between the infrastructure — identity management, privilege management, the things protecting your core and critical systems.

Ameya P: That’s where it collides. In 2026, understanding national sovereignty is critical. So much of that critical infrastructure is still controlled by Western companies. I know for a fact there is a government push to find domestic players and replace them where possible.

Ameya P: Wherever it’s customer-facing — customer experience, reaching customers faster, understanding usage patterns, building better products based on sentiment — AI gels fantastically. But the interconnect between heterogeneous systems, where you have custom integrators built to make those systems talk to each other — AI cannot gel there. You need a very stable integrator. The integration approach needs to work especially around security, authorizations, and ensuring tokens don’t outlast the handshake between two systems.

Ameya P: Those kinds of spaces are where AI isn’t going to gel well.

Pranav Manie: Makes sense. Having read your thoughts on the Citrini report — are you worried about India’s technological comparative advantage eroding? Is that something you think will happen in the AI age, or are you confident in India’s ability to build service and product companies in software? For context, we had another guest — in a proper live podcast, he came into the office — Pranay Kotasthane, a policymaker. His view was that India’s tech sovereignty will remain intact because a lot of smaller players will figure out a way.

Ameya P: I’m definitely optimistic. I engage with a lot of startups and new-age companies building natively out of India. Today it’s easy to go to the US and build something — and you do see a lot of Indian founders building things from the US. It’s easy to command a US premium and raise funds. But the people doing this out of India, out of sheer love of building something at home for their own people — that needs to be appreciated. And it’s happening.

Ameya P: That’s why I remain very optimistic. India’s tech canvas — which was primarily dominated by Infosys, TCS, and a handful of product companies — is going to change drastically over the next five years. We will see companies that are technologically at par, if not better, and delivery-wise, much cheaper. All of this happening with an India discount. So I don’t agree with the SES report in any way.

Ameya P: The one thing I do agree on is that the volume play is gone. Hiring 50,000 engineers a year, running them through six months of training, and deploying them on maintenance and testing — that model is not just threatened. It’s absolutely broken.

Ameya P: But this is going to bring quite a lot of change. The nature of jobs is changing. There will be some job loss — I’m not denying that. But new roles are emerging, like the ones I talked about — data labeling, data annotation, data pipeline and preparation workflows.

Ameya P: Prompt engineering itself is going to be a dedicated role. Someone who deeply understands supply chain and logistics, who’s been a functional consultant for 10 to 20 years and understands processes inside out — that person is well suited to create generic prompts that go into a prompt registry, which can be sold as a baseline feature to customers. You don’t want junior developers writing prompts at will and burning through your tokens.

Ameya P: So quite a lot of new roles are coming up. If you want to judge India’s technical sovereign capabilities, don’t look at the listed universe. Look where people are actually hands-down building something. That’s very encouraging. These industries will create more jobs.

Ameya P: I’ll give one more short piece of context, if you don’t mind.

Pranav Manie: Of course. Go ahead.

Ameya P: What happened is there was a tranche of old money — money that predated, I would say, 2015. That old money was very stable. It chased FDs, 12% ROI, and all that. And the new money chases innovation. It doesn’t just chase quarter-to-quarter analysis — it chases what you’re building, what the R&D spend is. And it wasn’t possible for old money to transition into new money just like that.

Ameya P: What happened is the owners of that old money grew old and passed their inheritance to the newer generation — our first-generation entrepreneurs. People like Achin, Albin, or even the Kamath brothers. These are, in my view, the first-generation entrepreneurs of the new money era.

Ameya P: They started building around 2015 or earlier. Over time they built companies, achieved scale. Whoever the initial investors were, they got their exits — Zomato listed and created quite a lot of capital to reinvest into something completely revolutionary.

Ameya P: We’re seeing what Pinar is doing right now. When these things happen over decades — because it doesn’t happen overnight — when old money transitions into new money and risk capital becomes available, more and more startups get funded.

Ameya P: And it’s so important that capital like this exists today, so people can at least try out whether something will work. Earlier, funding decisions happened on an Excel sheet based on your bottom line and projections.

Ameya P: That’s changing. And while the listed market is crashing, I don’t think there’s been a more exciting time for the startup ecosystem. I’m very positive about India turning the page.

Pranav Manie: Nice. Just one last question — it’s a theme that’s come up throughout this entire conversation, and it’s also partly why we started covering large cap and midcap IT companies: domain specialization. From the very start, when you talked about reinforcement learning, you said domain data is going to matter a lot. Palantir is a domain-specialized AI player, for instance. Do you think domain specialization is going to be a defining moat for Indian IT companies going forward?

Ameya P: Of course — not just for Indian IT, but for Indian businesses broadly. That’s actually what I’ve been writing about on Substack — India’s data moat. I’ve held this view for a long time: given India’s diverse footprint, our cultural diversity, the sheer depth and breadth of data we’ve generated is enormous.

Ameya P: Not just on the consumer side, but through Indian IT’s decades of work for Western clients. We still hold significant amounts of anonymized and pseudonymized data structures. We don’t hold the actual data — and it’s not legal to do so once an assignment is complete. But the learnings from that data, the structural patterns — that’s still with us.

Ameya P: And that’s why domain specialization matters even more today. The old pyramid erodes because execution gets automated. We’ve talked about outcome-based pricing, the importance of accountability, sovereign AI requiring local domain experts, the shift from capability to accountability — in my view, domain specialization is the answer to all of these things.

Ameya P: A company that deeply understands the tricks of the trade — not just the technology, but the regulatory nuances, the exception handling, the relationship between, say, a letter of credit and the underlying commercial transaction — or what happens when a shipment is delayed and a bank needs to make a call — those firms can build solutions that nobody coming from nowhere can replicate quickly. The knowledge required to build it correctly simply isn’t in any globally available training dataset.

Ameya P: That’s why, for all the startups coming up, I don’t see any breaking into the enterprise AI space.

Pranav Manie: That was our conversation with Ameya Pimpalgaonkar. We hope you enjoyed it. I took away plenty from it and came out a little wiser. We will be hosting more conversations with professionals across industries and disciplines soon. Until next time.