Some months ago, a senior manager at one of India’s large IT service firms described a conversation he had with a client who wanted to expand the engagement. More features, faster delivery, broader scope. The manager said ‘yes’ to all of it. Then he went back to his desk and realized he was not going to hire anyone to help do it.
He did not think this was remarkable. That was the most interesting part.
The latest results from the sector’s largest firms point toward this story, though reporting periods and currency effects differ.
HCLTech’s advanced AI revenue rose 62.1% year-on-year to $171 million, with record first-quarter deal wins of $2.4 billion, while it reduced headcount by 3,292.
Accenture, whose fiscal third quarter ended on 31 May, reported revenue of $18.7 billion, up about 6%, while its workforce grew only 1% year-on-year.
TCS’s revenue for the quarter ended June 2026 rose 3.2% year-on-year in constant currency, or 13.9% in reported rupee terms, but it added 9,279 employees on a net basis, its strongest quarterly headcount increase in four years.
IBM’s preliminary June-quarter figures missed expectations and produced a historic share-price decline, attributed to spending shifts, purchase timing, execution problems and delayed deal closures.
Others are yet to report results; Infosys’s for example, will be out on 23 July. But the IT sector’s direction of travel is visible.
For years, revenue and headcount in IT services moved together in a ratio planners could depend on. More work meant more people. The ratio was stable enough to anchor workforce planning, fresher hiring, pyramid structures and the model under which a young engineer in Bengaluru or Pune could join the industry, spend two years doing work no senior wanted to do and emerge with enough experience to do something more interesting.
That model assumed the existence of junior-level work.
The ratio may be coming apart. HCLTech grew revenue and shrank headcount in the same quarter. Accenture’s revenue grew several times faster than its workforce in percentage terms. What if these are not rounding errors, but structural changes becoming visible in the earnings trendline one quarter at a time?
The mechanism is not difficult to identify.
AI coding assistants have made the senior engineer significantly more productive. A developer who once needed a week to produce a substantial piece of work may now only need hours. The output looks the same; it passes tests. The pull request with new code arrives clean and well-structured. The ticket closes, the velocity metric rises and the sprint looks healthy. This productivity gain is no exaggeration.
What is harder to see is what has quietly been subtracted.
When a senior engineer produces in hours what once took a week, the work that disappeared was not all wasted time. Much of it was the process by which engineers understood what they were building. They typed every line, made choices about structure and approach, hit problems, resolved them, and carried a mental model of the system that was the residue of all that effort.
Junior engineers did the same at a slower pace, learning as they went. The code review that followed was two people comparing mental models. It was slower than it needed to be, but that slowness was doing something useful.
That process has been abbreviated and what also gets crunched is that process of understanding, not just the effort. The pull request that arrives in 40 minutes rather than four days has something missing that no metric captures. The reviewer who approves it, with very little time on hand, and who may well be using AI to check AI, is not comparing mental models.
Reviewers are examining a generated artefact and being asked to stand behind it as though the two were equivalent. Whether anyone fully understands what was built does not appear in anything these firms report.
IBM’s quarter illustrates a different consequence of the same transition. Clients are re-orienting technology budgets towards AI infrastructure and away from other categories. The economics of enterprise technology is shifting, even if not in the same way for every firm.
Large firms understand some of this. TCS added over 9,000 people in the June quarter and has been explicit about fresh hires as a policy. Eliminating the entry-level pipeline delivers short-term margin improvement but creates a structural problem five to ten years from now, when there would be nobody to elevate to senior engineer roles.
But the pipeline problem and the understanding problem are not the same thing. You can hire freshers and still produce code nobody fully comprehends, reviewed by engineers who trust the output because it looks trustworthy. A headcount may hold steady even as understanding declines.
The IT services industry is very good at explaining itself. Its quarterly calls are always polished, metrics are coherent and growth narratives are internally consistent. What the numbers do not show is whether the work being delivered is understood by the people delivering it, or simply produced by them.
That gap between production and understanding is not visible in any of these results. It will not appear in the next set either. But it is a reasonable bet that it will eventually appear, as these things do. Reality has a way of intruding on even the most coherent explanation.
The author is a technology consultant and venture capitalist.