What should I tell my kids to study?

Every parent of a teenager is asking some version of it. Which degree is still safe? Is coding finished? Should it be nursing, or plumbing, or something with people in it?

Behind the question sits an assumption: that somewhere on the list of subjects there is a right answer, and the job is to find it. It is worth taking that assumption apart, because the evidence of the last three years points somewhere else.

What is happening to first jobs

Start with what can be measured. Three Stanford economists tracked the payroll records of millions of American workers and looked at the youngest, those aged 22 to 25, in the occupations most exposed to AI. Between 2022 and mid-2025 their employment fell 16 per cent relative to everyone else, and by this summer the gap had widened to 19 per cent. Older workers in the same occupations were fine. Young workers in less exposed occupations were fine. The decline is concentrated precisely where careers begin.

Wages tell the same story. A study of 138 million American career profiles found that since late 2022, starting wages have fallen 4.5 per cent on average, and 6.3 per cent for junior positions, while senior pay held or rose. The door into working life is narrowing from the bottom.

The strange part

Here is what makes this different from every previous panic about machines: AI helps beginners most.

The best evidence comes from a study of 5,172 customer-support agents at a large software firm, one of the first careful field experiments with this technology. An AI assistant lifted output 15 per cent on average, but for the newest, least experienced workers the gain was 30 per cent. New hires with two months of experience performed like colleagues with more than six months. The machine compressed half a year of learning into weeks.

So the technology makes a junior employee more capable, and firms respond by hiring fewer of them, at lower pay. That sounds like a contradiction until you see it from the employer's chair. What the assistant really did was bottle the experience of the senior staff, everything the best workers had learned about handling a difficult call, and pour it into the newest ones. Once experience comes in a bottle, you buy less of it on the labour market. The gains from the bottling went to the firm. The people whose experience filled the bottle were not paid for it, and the juniors who would have earned while learning were simply not hired.

The ladder problem

Think of how anyone actually becomes good at a job. A master carpenter exists because a workshop once let an apprentice sand the boards. The sanding was never the point; it was the tuition. Nearly every skilled profession works this way: the tedious junior work is how judgement gets built, and junior wages are how society pays people while they build it.

AI does the sanding now, in law, in accounting, in software, in marketing. The work that trained the next generation is being done by machines trained on the last generation. The researchers behind the wage study put it plainly: entry-level grunt work is a crucial phase of learning, and without that pipeline it is hard to imagine how anyone develops the expertise for senior roles.

Notice what still commands a premium: the senior people. Their protection is tacit knowledge, the judgement that comes only from years of seeing things go wrong, which is exactly what the models have not yet absorbed. They are the last stronghold of the old wage curve, not an exception to its dismantling. A stronghold is a real thing. It is also, by definition, surrounded.

So what do you tell them?

First, the part that stays true. Judgement is the scarce thing, and it always was. Machines are extraordinary at producing answers and mediocre at knowing which answer matters, which is the skill built by studying anything deeply and honestly: mathematics or history, medicine or carpentry. Curiosity is a better guide than any list of safe professions, because the list will be obsolete before graduation. A person who can tell good work from plausible work will be employable longer than a person who merely produces it.

But then the harder part, which is the actual answer to the question. For fifty years, pay has been drifting away from work even for people with excellent skills; that story is told in another essay in this series. What is happening to your children's first jobs is the same drift, moving faster. If the bottom of the ladder pays less and holds fewer people while the machines' owners collect the difference, then the thing your children most need is not a cleverer choice of degree. It is a claim: savings that own a piece of the productive economy, a stake, however small, in the systems doing the work, and a country that builds such stakes for people who do not inherit them.

That last part is not something a seventeen-year-old can fix from a course catalogue, which is why it is unfair that the whole question lands on them. Tell them to study what they can love enough to be excellent at. Then ask the question their curriculum will not cover: not which job survives the machine, but who owns the machine?

---

Where the evidence lives. The studies in this essay are unpacked in chapter 3 of The Horse Is Here to Stay; the fifty-year story of pay drifting away from work is chapter 4. The running record of events is kept in the Stable.

One of a series of free guides answering the questions people are actually asking about AI and work. The book carries the evidence, the sources and the counter-arguments in full. These guides are published under a Creative Commons Attribution-ShareAlike licence: translate them, teach from them, quote them, build on them.