By BarathVector Editorial — 2026-08-13

The first Indian job lost to artificial intelligence may still have a person sitting in it.

The designation will survive. The salary may survive. The company may even report no layoffs. What disappears first is the bundle of junior work beneath the role: the first draft, the routine code fix, the document review, the research note, the reconciliation, the customer query. The employee remains, but the apprenticeship inside the job is hollowed out.

That distinction matters because the loudest AI employment debate is asking the wrong opening question. It asks how many workers will be fired. India should ask how many beginners will never be hired, and how those who are hired will learn enough to become experts.

The warning is in hiring, not unemployment

No serious evidence yet supports a declaration of AI-driven mass unemployment. India's own Economic Survey 2025–26 is careful on this point. Its analysis of US professional, business and information services found a post-2022 change in employment dynamics, including weaker responsiveness of employment to output growth. It expressly did not claim that generative AI caused the change.

That is the correct level of caution. Interest rates, post-pandemic hiring corrections, weak demand and ordinary business cycles can all suppress recruitment. Exposure to AI is not proof of displacement.

But the absence of mass unemployment does not clear the danger. Anthropic's March 2026 labour-market study, using US data and activity on its own systems, found no rise in unemployment among workers in the most AI-exposed occupations. It did find tentative evidence that the job-finding rate for people aged 22 to 25 entering highly exposed occupations had fallen 14 percent from its 2022 level. The result was only barely statistically significant, and the authors listed alternative explanations. It is not a verdict. It is an early alarm located exactly where an apprenticeship crisis would appear: entry.

India has more reason than most countries to listen. Its technology and business-services engine was built as a pyramid. Large junior cohorts performed codifiable work, learned under supervision and supplied the smaller layers of experienced staff above them. The model produced cost advantage for clients and mobility for Indian graduates. It also turned routine work into human capital.

AI can now perform much of that routine work. If the pyramid loses its base, the quarterly efficiency gain is obvious. The five-year talent loss is not.

Expertise is accumulated error

Senior judgment does not materialise at age 35. It is accumulated from low-stakes mistakes made at 23, corrected by somebody who has already made them.

A programmer learns architecture after years of debugging small failures. A lawyer learns strategy after reviewing ordinary contracts. An accountant learns fraud detection by reconciling dull ledgers. A reporter learns what matters after writing briefs that an editor cuts in half. The junior task looks replaceable when measured only by its output. Its second product—the trained human—is harder to see.

This is why simply telling graduates to "learn AI" is inadequate. A young worker can become faster with an AI system while receiving less exposure to the reasoning hidden inside the finished product. If the machine proposes the code, argument or analysis and the junior worker merely checks it, the worker may learn verification. But verification without construction can become pattern approval: accepting what looks plausible without possessing enough craft to locate the subtle failure.

The danger is not that nobody will work. It is that too few people will receive the sequence of responsibility through which competence is formed.

The productivity case is real

This warning should not be converted into nostalgia for inefficient work. AI can let a small Indian firm offer services it could not previously afford. It can help one capable employee do the work of a larger team, translate across languages and serve clients outside the metropolitan centres. The ILO and World Bank warn that developing economies may suffer disruption before receiving the productivity dividend; they do not deny the dividend.

Provider research points in the same direction. In Anthropic's June 2026 survey, large majorities of respondents reported gains in speed, scope and quality, and 68 percent said they learned more with AI. The sample consisted of Claude users, heavily overrepresented in technology and management, and the learning measure was self-reported. It cannot settle the labour-market question. It does show why workers and firms adopt the tools voluntarily: many experience genuine benefit.

The correct policy is therefore not to slow adoption to preserve yesterday's tasks. It is to make skill formation an explicit product of tomorrow's workplace.

A new apprenticeship compact

India needs an AI-era apprenticeship compact built around four decisions.

First, measure the missing rung. National employment statistics should publish fresher hiring, entry wages, training hours and progression in AI-exposed occupations separately from aggregate employment. A stable headcount can conceal a closed door.

Second, attach a talent obligation to public support. Companies receiving subsidised compute, public contracts or AI incentives should disclose how many early-career workers they hire and train—not as a permanent hiring quota, but as a visible return on public investment.

Third, create professional residencies. India has long used supervised training in medicine and chartered accountancy. Short, paid, assessed residencies can do the same for software, legal services, financial analysis, design and media. The state need not employ every resident. It can share the training cost while firms supply real work and named supervisors.

Fourth, redesign junior work instead of deleting it. Let AI handle repetition, but require beginners to make first attempts, explain decisions, inspect failures and carry limited responsibility that expands with demonstrated competence. A worker who only prompts never builds a profession. A worker who uses AI after learning the structure of the problem may build one faster.

The IMF's 2026 address on Indian AI placed the possible productivity gain at the centre of the opportunity. India should pursue it without apology. But productivity is output divided by input; a nation is not a quarterly spreadsheet. If today's gain consumes the process that creates tomorrow's experts, the arithmetic is incomplete.

The employment crisis will not necessarily arrive with a termination letter. It may arrive as an offer letter never sent, a first assignment never given and a generation asked to demonstrate experience nobody allowed it to acquire.

The job can disappear before the employee does. By the time the employee disappears too, the apprenticeship may already be gone.