Learning Nothing, Losing More

Every year, India’s big IT services firms put tens of thousands of new hires through six to ten weeks of training before letting them near a real project. The syllabus tries to cover everything — a messaging platform this week, a data warehousing tool the next, a bit of machine learning thrown in somewhere — and covers none of it well. Ask anyone who’s been through it what they actually remember, and the answer is usually: not much. Nobody inside these companies disputes this. It’s an open secret.

The real question isn’t whether the training works. It doesn’t, and everyone running it seems to know it — new hires still sit on the bench for months after “graduating,” no matter how well they did in class. The real question is why this ritual keeps happening, and what that says about the kind of company that needs it.

Most skills aren’t learned this way. Nobody teaches a carpenter six weeks of theory before handing them a saw — you learn by cutting real wood, badly at first, with someone experienced correcting you as you go. The same is true for most of what shows up in a corporate training syllabus. You get good at a tool by using it on a real problem, not by sitting through a tour of concepts that have nothing to hold onto yet. Most onboarding gets this backwards: it teaches applied skills like a university lecture, to people who will only actually learn the job later, on the job.

There’s one real exception, worth understanding because it shows what good training would look like. Enterprise software like SAP doesn’t just require knowing the tool — it encodes business logic a fresh graduate has never seen. Take a consignment sale: one company sells goods it doesn’t actually own, with its own rules for returns and stock reconciliation. You can’t make sense of how the software models that until you understand consignment as a business arrangement in the first place. That’s real training — not the software, but the logic underneath it. Done well, and done fast, someone comes out actually understanding why a company sets its systems up a certain way, not just which buttons to click.

But this is easy to lose. A course that starts out trying to teach real business logic can turn into the same grab-bag as everything else — a bit of this platform, a bit of that tool, some AI thrown in for good measure — assembled to look like training rather than to produce it. One company’s version leaves people able to explain why things are configured the way they are. Another’s leaves people able to recite a syllabus and nothing else. The exception was never really about SAP. It’s about whether anyone actually designed the course.

Part of the reason bad training survives is simple: the people designing it usually aren’t the people who understand the work. Courses get built by HR departments or outside consultants, borrowing templates meant for something else entirely — standardized instruction built for repetitive factory tasks, bolted onto knowledge work it was never meant for.

But there’s a deeper reason, and it has nothing to do with bad design. Product companies expect new hires to already know what they’re doing and let them pick up the rest by watching people who are better at it. There’s no reason to run someone through months of generic training when your revenue depends on the product, not the headcount. Staffing firms work differently. Their entire business is leasing out labour — billing clients for bodies across a shifting pile of technologies — and a training pipeline, however useless in practice, does real work for that model. It gives clients something to point to. It gives the firm a queue of people who look qualified enough to deploy. Whether anyone actually learned anything is beside the point. The ritual isn’t a mistake. It’s what the business actually is: a middleman matching labour supply to demand that increasingly sits somewhere else.

This matters because it’s exactly the entry-level layer — the freshers this whole pipeline exists to process — that’s most exposed to automation right now. A job whose core skill is thin by design, built on repeating well-understood tasks under supervision rather than real expertise, is also the easiest job for an AI system to fake convincingly. You can already see the nerves in the hiring numbers: several major employers have cut entry-level recruitment hard, not because AI has actually proven it can do the work at scale, but because they expect it to soon. That’s not a company reacting to a problem. That’s a company that no longer believes in its own training pipeline.

None of this — machines replacing routine labour — is new. What decides whether a wave of automation is merely painful or actually ruinous isn’t the job losses. It’s where the money that used to pay those wages ends up. When a factory automates inside a country, workers lose income, but the extra profit lands with an owner who still lives there, and mostly gets reinvested there too — into new ventures, more capital, demand that eventually works its way back through the economy. The distribution gets worse. But the money stays.

India’s IT services sector breaks that pattern. It exists to lease out labour to capital sitting somewhere else — mostly American. So when AI makes that labour cheaper to replace, the gains don’t move from a worker to a domestic employer. They go straight to the foreign company that no longer has to pay for the work. The loss here is real: wages gone, hiring frozen. The gain doesn’t show up here at all. That’s a worse deal than the usual story of automation — not an unequal split within one economy, but money leaving the country entirely.

This isn’t the first time. India’s experience of industrialisation looked nothing like the countries that used the same machines to build lasting wealth — not because of anything inherent, but because the institutions running the country at the time weren’t built to capture the gains. That gap never closed. China’s manufacturing dominance wasn’t just about cheap labour; it came from decades of production knowledge, real economies of scale, and market access that took years to build. Those advantages compound. Once manufacturing had settled somewhere else, no amount of trying harder was going to win it back on price.

The AI moment looks like the same trap wearing new clothes. India’s IT sector sits in almost the same spot in the global economy that its industrial base did fifty years ago — supplying labour to capital that lives elsewhere, absorbing the cost of every disruption without much claim on what it produces.

Part of why India hasn’t built its way out of this comes down to its own market. Most of the country doesn’t have much money to spend, which limits what companies can sell domestically and kills the case for building expensive, high-value industry in the first place. An economy whose main selling point is cheap labour is standing on thin ice — there’s nothing to fall back on once that advantage shrinks, or gets automated away.

Compare this to how Japan, and later China, actually built their industrial base. Both poured money into heavy infrastructure — rail, especially — during periods of real hardship, including actual food shortages. That spending wasn’t about welfare in the short term. It was about triggering a chain reaction: moving labour where it was needed, creating enough demand to justify building steel industries, kicking off cycles of growth that eventually paid for the welfare the initial spending had ignored.

India has never really been willing to make that kind of bet. Big infrastructure spending — high-speed rail is the obvious recent example — gets hammered publicly for taking money away from people who need food, not trains. Whatever the merit of that argument, it’s also exactly the argument that rules out the kind of investment that made other countries rich.

Underneath the budget argument is something less about money and more about instinct: a culture where being poor reads as honest, and being successful reads as suspicious. The idea that poor people are decent and rich people are probably crooked isn’t unique to India, but it runs deep here, and it has real consequences — it makes the kind of upfront, unequal, patient investment that built other countries’ industries almost impossible to sell politically, no matter how well it would pay off.

Which brings this back to where it started. The training nobody learns anything from and the industrial future the country keeps missing aren’t two separate stories. They’re the same story at different sizes. Both are about copying a shape — a curriculum, a rail line, a development plan — without the substance that made the original work. Both come from the same refusal to make a hard, upfront bet on a bigger payoff later. And without a real change in either how these companies train people or how the country invests, the pattern holds: the disruption arrives from outside, on someone else’s schedule, paid for with someone else’s money, and someone else ends up keeping most of what it makes.