AI and jobs

A teenager asks what they should study. A school wants subject choices signed off by the end of the month. You have to answer, and the honest answer — that nobody knows what that field looks like by the time they finish training for it — is no use at all to somebody filling in a form. So you answer anyway, with whatever you have, aware the entire time that the ground under the answer is moving.

What makes this a serious objection rather than free-floating anxiety is the second problem underneath it. The people telling you this technology will transform work are, in large part, the people selling it. A company saying its product can do the work of a department is describing a product, and the description is also a sales pitch. That does not make it false. It does mean the forecast cannot be taken at face value, and it means the loudest voices in the room have a direct interest in the answer.

There is a mirror version of the problem on the other side. The reassuring forecasts — that new work always appears, that this is one more wave like the ones before it — come from people with their own reasons to stay calm, and they are usually delivered with the same confidence and the same thin evidence. A parent looking for a straight answer is standing between two groups who are both very sure and who cannot both be right. Noticing that is not cynicism. It is an accurate read of where you are standing.

Start with what needs no forecast. Specific tasks inside many jobs are now being done by software that could not do them a few years ago: first-draft marketing copy, routine translation of routine documents, basic image and layout work, the first tier of customer support, and the more repetitive end of junior programming. These are not hypothetical capabilities. They are in use, they are cheap, and the output is frequently good enough for the purpose it is put to.

Some employers have responded by slowing or stopping junior hiring in exactly those areas. This shows up as roles that are never advertised rather than as redundancies, which makes it quieter and much harder to point at — a job that is not posted leaves no trace anywhere. It is described widely enough, and across enough different kinds of work, that treating it as anecdote would not be honest.

The pattern is uneven. It is not spread tidily across the economy; it is concentrated in work that is text-shaped, screen-shaped, and checkable — the kind of task a competent person can verify in a few minutes. Whole categories of work are barely touched. Which side of that line a particular job falls on is not obvious from the outside, and the line does not follow prestige, or pay, or how much schooling the job required.

The part that matters most for a parent is where the effect concentrates. The tasks being absorbed are disproportionately the ones handed to the newest person in the building: the first draft, the simple ticket, the routine translation, the small self-contained piece of code. That is not a coincidence. Those tasks go to beginners because they are contained and easy to check, and that is the same property that makes them easy to hand to software.

So the pressure lands on entry-level work, which is exactly where a young person starts. That should be said without cushioning it: if the shallow end of a profession is the part under pressure, the difficulty is not spread evenly across a career. It is concentrated at the point your child is standing at. A parent who reads this and feels the worry sharpen rather than ease is reading it correctly.

The most useful thing to hand a child here is probably not a career recommendation. It is the expectation that work changes: that most people do several different kinds of it, that the version someone trains for is rarely the version they retire from, and that this was already true for a great many people before any of this arrived. A young person who believes one decision at eighteen settles the question is fragile in a way that has nothing to do with which field they picked.

None of this makes the risk smaller. It is what is available to do while the risk is unresolved, which is not the same thing and should not be mistaken for it. You are not being offered a way to make the uncertainty go away. You are being offered ways to place a smaller bet on the parts nobody can see yet, and to make sure your child arrives carrying something that transfers.

Whether new work appears fast enough, and for the same people, is the whole question, and it is open. New occupations have followed earlier technologies. But “eventually, somewhere, for someone” is a different claim from “soon, nearby, for the person whose job went”, and the distance between those two has historically been measured in careers rather than in months. The people who fell into that gap were not compensated by the eventual recovery. There is no way to know from here which side of it your child lands on.

The apprenticeship path is the second unresolved thing, and it may be the more consequential one. If the tasks that used to train beginners are absorbed, a profession has to invent another way of producing experienced people or slowly stop producing them. Nobody has shown that a replacement works, because nobody has run one long enough to find out. Every argument you will read about this reasons forward from principle, on both sides, and that is as true of the optimistic ones as the bleak ones.

The third is distribution, and it is a decision rather than a fact. If the technology makes work more productive, the gain can go to wages, to lower prices, or to owners, and which of those happens is settled by bargaining power, law, and policy. Nothing in the software decides it. That is worth knowing precisely because it is usually discussed as though it were fixed in advance, by people who benefit from it being treated that way.

None of this resolves into advice. A parent who looks at it and concludes that the risk to their child’s field is real, and who plans conservatively as a result — a broader path, a fallback, a slower commitment, a trade or a training route that looks less exposed — has drawn a legitimate conclusion from the same unfinished picture everyone else is looking at. That is not pessimism, and it is not a failure to understand the technology. It is a defensible reading, and the reading is yours to make.