The End of India's IT Bargain

On the weekend of October 10 and 11, the Cockroach Janta Party took its argument to New Delhi, and New Delhi answered with barricades. Thousands of young demonstrators ran into police restrictions, cordoned avenues and detentions. Their leaders, Abhijeet Dipke, Saurav Das and Ashutosh Ranka, were taken into custody and later released. The demand this time was specific and procedural: Chief Election Commissioner Gyanesh Kumar should resign over revisions to the voter rolls.

The movement’s name is a captured insult. It began in May 2026, after the Chief Justice of India made a remark about unemployed youth that reached for the image of cockroaches. Students took the word, printed it on placards and built a party around it. Over the summer they collected results that most protest movements only claim. The biggest was the resignation of an education minister. Their website still puts exam reform next to electoral integrity, which shows how the movement sees itself: first a fight over who gets a fair shot at the test, now a fight over who gets counted.

These are arguments about access, fairness and representation — whose paper is marked honestly, and whose name stays on the roll. Nothing in the movement’s stated demands concerns artificial intelligence, and nobody should put that motive in its mouth. Still, the timing deserves attention. A generation is fighting hard for a fair place in the competition just as the economics at the end of that competition — the technology job that turned engineering degrees into middle-class lives — are being rewritten.

For three decades, India sold the world human capacity to produce software

The bargain was never written down, but every middle-class Indian household could recite it. A child studied mathematics and physics with monastic intensity and sat the entrance examinations. If the child cleared the Joint Entrance Examination, there was an Indian Institute of Technology. If not, there was one of the private engineering colleges that spread across Tamil Nadu, Andhra Pradesh, Karnataka and Maharashtra to meet demand. At the end of four years came campus placement: an offer letter from Tata Consultancy Services, Infosys, Wipro, HCLTech or Tech Mahindra, a training stint, a desk in a glass building, and a salary that could repay the education loan within a few years.

Above that floor sat a ceiling worth reaching for. The offshore delivery model sent work from New Jersey and Frankfurt to Pune and Chennai, and the H-1B visa sent some of the people the other way — to client sites in Texas and California, to green cards, to remittances that bought flats back home. Families mortgaged land, paid for coaching centres and postponed their own comforts, and they were not deluded. For millions of households the arrangement worked. It was the most reliable path to prosperity that liberalised India offered anyone without inherited capital.

Look at what was actually being sold. Indian IT exported a capacity: thousands of trained people who could be assembled into teams, billed by the hour or the project, and scaled up when a client’s ambitions grew. Revenue followed headcount. The industry’s famous pyramid had many junior engineers at the base and a thinner layer of managers and architects above. It was a business model, and also a social contract with every family that sent a child to engineering college.

AI threatens the link between headcount and output, not the appetite for code

The lazy version of this story says AI will replace programmers. That version is easy to dismiss, because every previous wave of automation, from compilers to cloud platforms, ended with more software and more people writing it. The more precise version is more unsettling for an industry that employs millions.

Traditional outsourcing is priced in human effort. A bank wants its core systems modernised, so a vendor proposes a team of a certain size for a certain number of months. Even fixed-price contracts are estimated by staffing a pyramid and adding margin. The client buys software but pays, in effect, for the number of people needed to produce it.

Agentic coding tools change that arithmetic. A smaller team of experienced engineers can now direct software agents that generate code, write and run tests, read the failures and iterate, with humans reviewing, correcting and deciding. The bank still needs its systems modernised and may want more modernisation than before, because each unit of it has become cheaper. What it may no longer need is the same number of external staff to get there.

Demand for software could grow while demand for software labour falls.

That is the destabilising change. Indian IT has survived many technology shifts by retraining its workforce and selling the next service: mainframes to client-server, client-server to web, web to cloud. Each transition kept the core equation — more work for more people. This one threatens the equation itself.

The largest Indian IT employers are reorganising, not collapsing

The first evidence is in the payroll data, and it is suggestive rather than dramatic. In fiscal 2026, the combined workforce of TCS, Infosys, HCLTech, Wipro and Tech Mahindra shrank by roughly 7,389 people, according to Moneycontrol and IANS. That reversed the previous year’s expansion. TCS, the industry’s largest employer and its bellwether, announced a reduction of about 12,000 roles.

The picture is not one-way. In the September 2026 quarter, TCS added 4,258 employees and took its headcount to about 598,056. It is still hiring graduates, with about 24,000 brought in so far this year against a target of 25,000. It also plans roughly 8,900 forward-deployed engineers — specialists embedded with clients to integrate enterprise AI into their operations — as the Financial Express reported on October 9, 2026.

This does not look like an industry falling off a cliff. It looks like an industry changing what it sells. The new prize is revenue per employee, not total headcount. The new hire is the engineer who can sit inside a client’s business and wire agents into its workflows, not a cohort of freshers waiting to be deployed onto maintenance contracts. The firms are betting they can grow by selling output and expertise rather than bodies.

Whether that bet pays the workers depends on who captures the productivity gain, and there are four broad possibilities.

Outcome	IT services companies	Indian workers
Providers retain the gains	Margins widen as smaller teams deliver the same contracted work	Fewer roles per contract; rewards concentrate in senior and specialist staff
Customers capture the gains through lower prices	Revenue per project falls; firms must win more projects to stand still	Pressure on wages and hiring as vendors protect margins
AI creates enough new demand	Volume rises fast enough to offset smaller teams per project	Employment can hold or grow, but with a different skills mix
Customers internalise the work with agents	Outsourcing shrinks as clients build in-house with small teams and AI	The deepest threat: the external pipeline that absorbed graduates contracts

The last row is the most dangerous for India. The first three are fights over how to split a pie. In the fourth, the client stops ordering from the bakery. If a Fortune 500 company decides it can modernise its own systems with a lean internal team and a fleet of agents, the problem for Indian IT is no longer pricing. It is relevance. The same pattern is already showing up in The SaaS Recoil, where companies weigh building their own software against renting it.

The H-1B channel is narrowing at the same time, for reasons unrelated to AI

The second pillar of the bargain was mobility, and it is under pressure from a different direction. A Federal Register document published on September 23, 2026, in connection with Proclamation 11069, reported that combined H-1B registrations from the largest IT staffing and outsourcing firms fell from 24,946 to 2,055 — a decline of 92 percent. It also described a shift in the applicant pool toward higher-paid and more highly qualified candidates.

Attribution matters here. The document links the change to the $100,000 payment requirement and to wage-weighted selection, not to artificial intelligence. American policy, not American software agents, made the old high-volume visa strategy uneconomic. The cost logic is now spreading to renewals as well, as covered in The 50-50 Fee Now Follows the Extension.

The two pressures are separate, but they land on the same structure at the same moment. Offshore delivery is being squeezed because AI changes how much labour a unit of work requires. Onshore mobility is being squeezed because Washington has made the visa more selective and more expensive. For decades those two channels together defined the Indian talent-export model: do the work from India, or move to where the work is.

The open question is what becomes of that model when the importing economies need fewer people for the same technical work, whether those people sit in Bengaluru or Boston. The question goes beyond visas or vendors to the market value of a skill set that India manufactured at enormous scale — the dynamic explored in When Expertise Gets Cheap.

The speed of displacement is a scenario to test, not a forecast

Enterprise technology rarely changes gradually. It tends to move through long periods of pilots, proofs of concept and procurement committees, followed by an abrupt reorganisation once one large client proves the new model and its competitors follow. If AI-driven restructuring of IT staffing follows that pattern, the quiet of the past two years may be the preparation phase rather than the whole story.

Three variables will set the pace. The first is technical capability: whether agents can work reliably across the sprawling, poorly documented, decades-old systems that make up most enterprise IT, not just tidy new codebases. The second is economic advantage after all the costs are counted — including human supervision, security review, compliance, and recovery when an agent confidently breaks something in production. The third is organisational willingness: whether large clients will actually rewrite procurement rules, vendor contracts and governance frameworks built around billable teams.

None of these is settled, so this should be treated as a hypothesis with a clear test: watch whether revenue and headcount diverge. If TCS, Infosys and their peers keep growing revenue while employing fewer people, the labour economics of the industry have changed structurally even if no company ever announces it. If revenue later starts to fall as clients bring the work in-house, the challenge reaches past staffing to the outsourcing model itself.

An Indian IT company could become more profitable while employing fewer Indians

Here the economics become politics. If the industry produces the same revenue with fewer employees, the gains will go somewhere: to shareholders through margins, to the senior specialists and forward-deployed engineers who command premium salaries, and to customers through lower prices. It is far from clear that they will reach the broad base of graduates the pyramid used to absorb. The national champion of Indian upward mobility could post record profits and a shrinking payroll in the same year, and both numbers would be accurate.

When a private escalator slows, people turn to the state. Government becomes the institution of last resort for funding education whose returns are less certain, for creating jobs the private sector no longer generates automatically, for retraining at scale, and for social protection for the families who financed the old bargain. In that world, political legitimacy is no longer an abstraction. It becomes an economic asset, because a state that is trusted can ask for patience and a state that is not trusted cannot.

That is why the fights over examinations and voter rolls matter beyond themselves. The Cockroach protesters have not said AI is their grievance, and this article does not claim it is. But a generation that insists on fair tests and honest electoral rolls is demanding the institutional trust that any transition will draw on.

For decades, India’s students were told that education was the surest way out of poverty, that engineering was the safest profession, and that the world’s appetite for technical talent would keep growing. They competed. Families sacrificed. Universities expanded. Outsourcing companies built global businesses around that labour.

Now the economics are changing. AI does not need an entrance examination, a university placement or an H-1B visa. It does not need to emigrate to serve an American customer. With every improvement in autonomous software development, the human labour required per unit of technical output may decline.

The students began by demanding fair examinations, then turned toward the institutions that govern their future. The hardest question may be one that neither an education minister nor an election commissioner can answer:

What happens when a generation wins the right to compete, only to discover that the competition no longer leads where it once did?