The Economist published a piece this week examining why AI has not yet disrupted India’s IT outsourcing industry, a sector they treat as representative of the global market’s exposure to AI displacement. The conclusion: legacy code is messy, clients overestimate AI’s readiness, headcount keeps growing, and Nasscom expects its members to post combined revenue north of $315 billion this year. Crisis averted.
The analysis is not wrong. But it mistakes a lagging scorecard for a forward indicator.
The Brownfield Defense
The strongest argument from India’s IT executives is the brownfield one. The Economist quotes Atul Soneja, Tech Mahindra’s COO, distinguishing between greenfield environments (new systems with clean architecture, where AI excels) and brownfield ones, where legacy code, missing documentation, and interdependent services make AI deployment far harder. His argument, essentially: AI works great on a blank canvas, but enterprise reality is never a blank canvas.
Fair enough. Anyone who has tried knows the difference. And knows he’s correct.
But the defense assumes AI capabilities are static. They are not. The reason brownfield environments resist AI today is that the tools cannot hold enough of the system in view at once. They lose track of how services connect, where the undocumented dependencies live, how a change in one module cascades through twelve others. That limitation is dissolving. The context windows that govern how much code and documentation an AI model can reason over simultaneously have expanded dramatically in the past year alone. In practical terms, an AI that could previously read and reason over a few dozen pages of code can now process the equivalent of several thousand pages in a single pass. Agentic tooling, AI that can navigate codebases, run tests, and trace dependencies autonomously, is maturing in parallel.
The moat around brownfield complexity is real today. It is also eroding, month by month, release by release. And “yet” is carrying enormous weight in a $315 billion survival thesis.
Revenue Is a Lagging Indicator
The Economist cites slightly-better-than-expected quarterly results and rising headcount as evidence of resilience. But aggregate revenue growth reflects contracts signed twelve to eighteen months ago. It does not tell you what is being signed today. As we wrote in Vendors Under AI Pressure, when revenue compresses, margins tighten, and the pressure shows up in delivery long before it shows up in earnings.
The Consulting Pivot and Its Limits
The piece quotes Nandan Nilekani, one of Infosys’s founders, projecting that AI-related services could be worth $300 to $400 billion by 2030. The instinct is right. Companies need help deploying AI effectively. Understanding organizational context, business logic, integration constraints: that is consulting work. Humans remain better at it. This is the strategic layer where experience and incentive awareness still matter more than automation.
The limitation is scale. India’s IT industry employs roughly 5.4 million people. The vast majority are not performing strategic consulting. They are writing routine code, maintaining test suites, handling support tickets, processing data. The strategic consulting play can preserve the industry’s margins. It cannot preserve its headcount.
This is the distinction the optimistic narrative elides. The services that survive AI pressure are not the services that employ the most people. For buyers navigating this shift, the question becomes how to select an AI partner that understands these structural dynamics, not just the technology.
Context Windows Will Prove This Thesis Wrong
The Economist closes by observing that AI’s effect on the sector remains “unclear and uneven.” That framing lets you be right regardless of what happens next.
Here is a more specific claim: context window expansion alone will unravel the brownfield defense. When an AI agent can ingest an entire legacy codebase, every module, every configuration file, every undocumented dependency, and reason over it coherently, the argument that these systems are too messy for automation stops holding. That capability is not speculative. It is the current trajectory, and it is accelerating. We have been writing about this speed since we launched.
The industry will not collapse. It will compress. When one skilled developer with the right tooling begins doing the work of several, labor arbitrage narrows. Not overnight. Gradually, and then structurally.
The disruption has not arrived. That is not the same as saying it will not.
Photo of brown field by the author, traveling through Spain with his dad.