It is late August in Dhanbad. The placement notice on the college board says the company that has come every year for fourteen years will not be coming this year. Nobody has explained why, exactly. The seniors who landed the job last year do not return calls, except one, who says her team has been told to do the work of three.
This is most engineering campuses in tier-two India this year. The pyramid that used to absorb a hundred thousand fresh graduates a year is no longer that shape.
It is not a recession. The companies are still making money. Many of them are making more money than before. They have just stopped needing the same number of you.
You know why. We have all known why for two years. The work that used to take a fresher six months to learn is now done by software in six minutes. The honest people in the IT industry will tell you this if you catch them after the third drink. The dishonest people will tell you the offshore model is still alive and well.
You are twenty-three. You have a degree from a college nobody outside your district has heard of. The advice you are getting from everyone, on every reel, is “learn AI.” None of them tell you what that means.
We will try.
What is actually happening
The Indian IT services industry was built on a deal. American and European companies had work that required a lot of people. Indians had a lot of people, decent English, and clean enough engineering colleges to produce them at scale. The deal made many fortunes and lifted tens of millions of households into the middle class.
The deal worked because the work was process work. It was the same kind of task, repeated, by junior people, supervised by senior people, billed by the hour. Tickets, code reviews, support calls, data cleanup, basic coding from a spec. The pyramid was made of process work at the bottom.
Process work is exactly what the new models are good at. They are not better than your best engineer. They are vastly better than your fiftieth engineer. The bottom of the pyramid is, in the language of the boardroom, “rationalising.”
The companies will not say this on the earnings call. They will say things about productivity. The substance is the same. They are hiring fewer juniors, asking each junior to do more, and slowly raising the bar for what counts as junior at all.
On the brighter side, the GCC wave is real. Western banks and tech firms are setting up Global Capability Centres in Bengaluru, Hyderabad, Pune, Chennai. They are hiring. They are paying more than the services companies ever did. But they are hiring people who can do real product work, not people who can grind through tickets.
So the floor is rising. The question is how to get over the new floor.
The lie called “learn AI”
The advice “learn AI” is the most repeated and least useful thing being said to your generation right now.
It is not advice. It is a wave of the hand from people who have already made it.
What does “learn AI” actually mean for somebody twenty-three years old with a B.Tech from a college nobody knows? You cannot become an AI researcher in three months. The PhDs at OpenAI and DeepMind are not your competition. You should not pretend they are.
Three things are within reach.
First, get very good at building things with AI, not at training it. Pick a real problem somebody you know has. Your father’s small business has it. Your aunt’s clinic has it. Your panchayat has it. Build a thing that solves it using the tools that already exist. A WhatsApp bot. A scheduling app. A small dashboard. The skill is in the building and the user understanding, not the model. The model is now a commodity. The product around it is not.
Second, own one Indian vertical the models do not understand well. Indian tax, Indian property law, Indian agriculture, the way a small kirana store actually does its accounts, the back-office of a Tier-3 hospital, the operations of a school in Bihar. The internet has terabytes of American legal text. It has very little of any of these. Whoever knows the vertical, plus knows how to wrap an AI around it, owns the next ten years of that vertical.
Third, build in your language, for the people who do not speak English. The next half-billion Indians coming online are not going to type queries in English. They are going to talk to their phones in Bhojpuri and Marathi and Tamil. The voice models are good enough now. The products are not built. The market is open.
These three are not safe. They are not what your parents will understand when you say it at the dinner table. They are real.
The deeper move India still has not made
For thirty years, India has been the world’s back office. The next thirty years will not reward back offices. They will reward people who own products.
A back office sells hours. A product sells outcomes. AI eats hours. It does not eat outcomes.
The companies in India that will matter in this next age are companies that build something a real person uses, not companies that supply hours to companies that build something. There are a few of them already. The digital payments rail. A handful of SaaS exporters. The public-good stack like Aadhaar and UPI. There need to be a thousand more.
For an Indian under thirty, this is not bad news. This is the door opening, finally, to the kind of work that builds wealth instead of trading time for it.
But the door is narrow. It is going to ask you to do work nobody assigned. To learn things nobody is paying you to learn. To put a year of evenings into a project that may not pay for two years and may never pay at all.
Most people will not. A few will. The ones who do will be the founders of the next decade.
What you can do this month
Pick one problem somebody real has. Spend ten hours building the worst possible version of a tool that helps them. Show it to them. Watch them not use it. Fix it. Show it again.
You will learn more in those forty hours than in two semesters of any course called “AI for Everyone.”
The pyramid you were promised is gone. The thing on the other side is bigger, and it has fewer guards at the door than any economy India has ever had.
Walk in.
