Three findings about artificial intelligence and work have been circulating for most of this year, and they are usually presented as a contradiction.
Employment among 22- to 25-year-olds in the occupations most exposed to AI — software development, customer service and similar — is down by something like 16 percent. Companies adopting AI have cut hiring of junior staff by around 13 percent. And there is no detectable rise in aggregate unemployment among AI-exposed workers since late 2022.
The third is regularly used to dismiss the first two. It should not be. All three can hold at once, and the reason they can is a property of the instrument rather than of the labour market.
What unemployment counts
To be unemployed in the official sense you must be out of work, available for work, and have actively looked for it recently. It is a status people arrive at, overwhelmingly, by leaving a job they had.
A firm that lays off a thousand people creates a thousand candidates for that status, and the rate moves.
A firm that decides not to open forty graduate positions this year creates none. The forty people who would have filled them are somewhere else — in school, in a job below their qualifications, in a role in a sector that is not exposed, or in the count as unemployed for reasons that will be attributed to something other than a posting that never existed. Nobody is displaced. There is no separation to record. The vacancy is not withdrawn, because it was never advertised.
This is not a flaw anybody hid. The unemployment rate has always measured the stock of people looking for work, and it was designed in an era when the thing that happened to workers was that jobs went away. It measures that superbly. It was never asked to detect a cohort that is quietly not being let in.
Why the entry level first
The pattern in the data is specific to junior roles, and there is a straightforward reason that has nothing to do with how clever the technology is.
Entry-level white-collar work is disproportionately the well-specified part of a job: the research memo, the first-pass ticket triage, the routine query, the document review, the code that is annoying rather than hard. It is the portion of professional work that comes with a clear brief and a checkable output — which is exactly the portion that a system can attempt, and exactly the portion where a manager can tell whether the attempt worked.
That is the same boundary this desk found when companies could not tell whether their AI worked at all: the tasks organisations automate confidently are the ones where success is legible. Junior work is legible almost by definition. That is what makes it junior.
The consequence is that the first thing an organisation stops buying is not labour in general. It is the specific tranche of labour that used to be how people entered the profession.
The part that will not show up for a decade
Cutting junior hiring is defensible on this year's numbers. Junior staff are a net cost for a period before they are productive, and if a system covers the output, the arithmetic in the first year is favourable.
The problem is that the junior years are not only production. They are how a firm manufactures its own seniors, and there is no external market that supplies people who have already done ten years at your company. The cheapest recruiting channel any organisation has is the one that moves people it already employs, and it depends entirely on having hired them at some earlier point.
A firm that stops taking juniors for five years has not saved five years of salary. It has removed five cohorts from the pipeline that produces the people who will run it, and that shortage arrives on a delay long enough that nobody currently making the decision will still be in post to explain it.
The sentiment number is the one to sit with
Gallup, surveying for the Walton Family Foundation in February and March, found Gen Z excitement about AI down from 36 percent to 22 in a year, and the share reporting anger up from 22 to 31.
Read next to the hiring data, that is not a survey about technology. It is a survey about a labour market in which the entry point has narrowed, conducted among the people standing at it. Whether they are correct about the cause is a separate question, and honestly one nobody can yet answer with confidence. What they are describing is consistent with what the employment numbers show for their age group in their occupations.
What to watch
Not the unemployment rate, for the reason this piece is about.
Watch the hiring rate for workers aged 20 to 24 in professional and business services, and watch the share of job postings requiring three or more years of experience. Both are flow measures and both would register a door closing, which the stock measure cannot.
And watch whether any large employer publicly commits to a graduate intake number for 2027 and 2028. Under-hiring juniors is individually rational and collectively expensive, which makes it exactly the kind of decision no firm changes on its own. If anyone breaks first, it will be a company that has already run out of people to promote.
The finding of an approximately 16 percent fall in employment among workers aged 22 to 25 in AI-exposed occupations such as software development and customer service; the reported reduction of about 13 percent in junior hiring at US companies adopting AI; and the absence of a detectable rise in aggregate unemployment among AI-exposed workers since late 2022 are as reported in research summarised across labour-economics coverage during 2026, principally work associated with Stanford's Digital Economy Lab. The Gallup survey conducted for the Walton Family Foundation in February and March 2026, showing Gen Z excitement about AI falling from 36 to 22 percent and anger rising from 22 to 31 percent, is as published by Gallup. Projections that AI could eliminate between 10 and 50 percent of entry-level white-collar roles are forecasts made by others, are not treated here as findings, and are attributed where mentioned. The argument about what unemployment statistics measure is our own.
Topics ailabourhiringlabor day
