Every large Indian IT firm has a pyramid. At its base, fresh graduates hired by the thousand from the NITs, IITs, and the vast middle tier of engineering colleges, onboarded in batches, trained in six weeks, and deployed to write code that is reviewed, tested, and shipped by the layers above them. In the middle, experienced engineers managing projects, bridging the gap between client requirements and junior output. At the top, a thin layer of architects, account managers, and domain experts whose judgement the whole structure depends on. This pyramid, roughly 70% junior, 20% mid-level, and 10% senior, has been the defining feature of Indian IT for three decades. It is also, now, a liability.
The threat is not abstract. Generative AI handles syntactic correctness, the ability to produce code that compiles, runs, and conforms to structural patterns, with increasing reliability as recent model announcements from Anthropic and OpenAI show. The next wave of coding assistants, trained not just on repositories but on developer decision-making itself, is closing in on functional correctness: code that does what it is intended to do. When that frontier falls, as it is falling, the work that occupies the base of the pyramid becomes work that a model does for a fraction of the cost of a junior engineer’s salary. The base does not thin. It hollows.
What replaces it is not a smaller version of the same shape. The skills in demand in an AI-integrated enterprise are verification, architectural judgement, and accountability, skills that live near the top of the old pyramid and are, by definition, not scalable through campus hiring. You cannot train a thousand fresh graduates into the kind of judgement that catches a flawed AI-generated assumption embedded six layers deep in a financial model. That takes years of domain experience and the confidence to push back. These are senior skills. The new pyramid needs to be built around them.
What does this mean in numbers? Consider a notional delivery team of 100 people. Under the old model, 70 juniors cost roughly $15,000 each in fully loaded annual terms, generating a salary base of just over $1 million. Twenty mid-level engineers at $35,000 — add $700,000; 10 seniors at $80,000 — add $800,000. Total people cost: $2.5 million. Average cost per head: $25,000. This is the arithmetic of the old arbitrage.
Now invert. A restructured team carries 20 juniors, 50 mid-level AI supervisors and integration specialists, and 30 senior architects and model auditors. Juniors remain at $15,000. But mid-level roles now command $50,000 because AI supervision, prompt engineering, and integration work require genuine skill and bear genuine consequences. Senior roles now move to $100,000, reflecting the premium on judgement and accountability. The same 100-person team now costs $5.8 million. Average cost per head: $58,000. Personnel costs have more than doubled.
The saving grace, and it is a real one, is that the team need not be the same size. If AI tools genuinely absorb 40% of the coding work previously done by junior staff, the same output can be delivered by 60 people rather 100. A 60-person team structured on the new ratios costs $3.5 million dollars against the old $2.5 million. That is a net cost increase of roughly 36%, painful but survivable, and potentially offset by the higher billing rates that verified, accountable AI-integrated delivery ought to command.
I can make this example even more arithmetic-driven if I were to take in offshore/onsite splits on teams, but I will not venture there since the basic thrust of the argument does not change. But just to tease your minds, the offshore/onsite split has moved in recent years from 70/30 to 90/10, meaning most of the work is now done offshore.
The harder problem is not arithmetic. It is structural. Indian IT’s relationship with the engineering college system is built on volume. Infosys, Wipro, and TCS collectively hire tens of thousands of graduates a year. That pipeline is the implicit bargain between the industry and the institutions that feed it: we will take your output, train it, and employ it. A restructured pyramid absorbs perhaps a fifth of that volume. What the industry tells the NITs and the second-tier colleges when the offer letters dry up is a question no one is yet willing to answer publicly, but the answer will have to come.
The client side will adjust too, though not without friction. For years, enterprises have accepted the Indian IT pricing model precisely because the pyramid made it cheap. A delivery team that costs 36% more per engagement will face resistance, even if the output is more reliable. Indian IT firms will need to make the case not just that the new model costs more, but that it is worth more: fewer errors, faster recovery, verifiable correctness, and a human who can explain what the system did and why. Plus, this will need a different offshore/onsite mix; more people presumably onsite with higher visa costs. But trust, as I have argued before, is becoming the only genuinely billable commodity.
The firms that survive this transition will not be those that resist the inversion. They will be those who manage it deliberately, shrinking the base at a controlled pace while building the senior layer that the new model demands. The ones that move too slowly will find the base collapsing under them before the senior layer exists to catch the weight.
The pyramid is not going away. But the version that built this industry will not be the version that sustains it.
Disclaimer: The views expressed are the author’s own and do not reflect the official policy or position of Financial Express.