What to study if AI scares you

‘Fast forward frugal swadeshi enabled AI-powered super computing technology’

A seventeen-year-old asked me recently whether there was any point in choosing a course at all. Whatever she studied, she said, would be obsolete before she finished it. Her parents were in the room, and they were nodding.

The fear is genuine, and it is not stupid. It is also built on a misreading of what the technology in question actually does.

Economists who study this properly do not ask whether a machine can replace a job. They break the job into tasks and ask which of those tasks a machine can do. The honest answer for most white-collar work is a large minority: perhaps a third, in some roles a little more. What follows from that is the part almost nobody says aloud. If a third of the work becomes cheap, the remaining two-thirds becomes more valuable, not less, because it is now the scarce ingredient. The people who lose are not the ones whose jobs contained automatable tasks. They are the ones who kept doing those tasks by hand while someone else moved on.

Software engineering is the clearest evidence available, because it is the field the technology has hit hardest and earliest. If the collapse thesis were right, we would see it there first. Instead, competent engineers are busier than they have been in years and the shape of the work has changed. Specialists who once did only front-end or only back-end now do both. The same broadening is showing up elsewhere. A marketing coordinator who once ran one stage of a campaign now runs the whole cycle. A recruiter who once sourced candidates now handles the process end to end. Some teams have started asking marketing candidates what software they have built, and expecting an answer. Finance staff are writing their own scripts to open documents, check figures against each other and flag what needs a human eye. Recruiting departments are hiring engineers to sit inside them.

The bargain is visible in that list. Tasks come off your plate, and in exchange you are expected to be good at more things. That is a demand for more education, not less.

Which brings me to the part students most need to hear, and it is not about employment at all.

The evidence now emerging is that when students use these tools on their coursework, their marks go up and their retention goes down. They score better on the assignment and worse weeks later, when the assignment is gone and only what they absorbed remains. This is not mysterious. Getting an answer and learning something are different activities, and one of them has been made almost frictionless while the other has not. Anyone who works with these tools daily can confirm it privately: you solve a problem, ship the thing, and six months later you cannot reconstruct how it worked, so you ask again.

So the practical rule is to separate the two modes and be ruthless about it. When you are producing, offload whatever you like. When you are learning, do not. The struggle is not an inefficiency in the process; it is the process. A student who cannot tell which mode they are in will graduate with a good transcript and nothing underneath it.

This matters because of where the durable human advantage sits. It is not raw intelligence. It is context: the thousands of small things you know from having been in the room, watched something fail, seen a customer’s face change, absorbed how your field actually behaves as opposed to how it is described. Judgement and taste are what we call context after it has been compressed by experience. No shortcut to it exists, which is precisely why it will remain valuable, and why sitting through a demanding degree is worth more now than it was five years ago rather than less.

There is a related trap worth naming. Using these tools every day is not the same as being good at them. Firms that assess this at scale report that most employees badly overrate their own level. Opening a chatbot each morning to tidy emails is adoption. Knowing how to structure a prompt, chain steps together, feed a model your own working documents and judge when its output is confidently wrong is proficiency, and the gap between the two is enormous.

So, what should the seventeen-year-old study?

There is no safe major, and anyone offering one is selling something. Choose a field with real intellectual depth, ideally one you can stand to spend years inside. Learn to build things, whatever your discipline, because the cost of building has fallen so far that the constraint is now deciding what is worth building. And cultivate the quality employers are quietly listing more and more often, which is agency: noticing a problem and doing something about it without being told.

One warning specific to us. University curricula in India are revised on a timescale of years, through committees and approvals, while this field moves every few months. A great many students are being prepared for the jobs of 2022. Work hard in your classes anyway, then close the gap yourself with free courses and small projects built for real users.

Fear makes people freeze, and freezing is the only strategy here that reliably fails. The skill that protects you is not any single tool. It is how fast you can learn the next thing.

The author is an entrepreneur, public policy commentator and columnist; Views presented are personal.