There are two loud stories about AI and jobs. In one, a wave of automation is about to wipe out white-collar work. In the other, nothing is happening and the panic is a product of people with something to sell. Both are told with great confidence and very little data.
There is a third story, and it comes from payroll records. It says that almost nobody has lost a job to AI, and that one specific group is quietly being left behind anyway. If you are a senior engineer, it explains why your job feels safe. If you are trying to become one, it explains why the door seems harder to open.
Where the data come from
In August 2025, three economists at Stanford’s Digital Economy Lab published a paper with a memorable title, “Canaries in the Coal Mine?”. They used payroll records from the payroll company ADP and compared occupations by how exposed their tasks are to generative AI.
In August 2026 they updated it with data through June 2026. The update is the best current snapshot of what is measurably happening, and its first finding is the one that gets left out of most headlines: we do not see widespread, economy-wide job displacement associated with AI.
The gap
The second finding is the canary. Employment among workers aged 22 to 25 in highly AI-exposed occupations now stands about 19% below where it would be if it had kept pace with similarly aged workers in less-exposed occupations.
Three details give that number its shape.
It is growing. By the same measure, the shortfall was 15% at the July 2025 data point. A year of additional data made the gap wider, not narrower.
It is about age, not occupation. Experienced workers show no comparable gap. A 40-year-old in the same exposed occupation is doing fine.
It is happening at the door, not the exit. The authors report that it operates primarily through reduced hiring of young workers rather than increased separations. Young people are not being fired. They are not being taken on.
The update also reports that declines are concentrated in occupations where AI usage substitutes for human tasks, while uses that complement human work go with flat or rising employment.
Why juniors and not seniors?
The authors’ proposed explanation is about two kinds of knowledge. Some of what a job requires is codified: the material in textbooks, documentation and training courses. The rest is tacit: judgement built from experience, knowing which corner of a codebase is fragile, sensing when a requirement is wrong.
Language models are trained on codified knowledge, which is exactly what a new graduate brings. They have much less of the tacit kind, which is what a senior person is paid for. On that reading, AI competes directly with the newcomer and makes the veteran more productive. The paper reports the declines for young workers are concentrated in occupations that involve codified knowledge.
It is a tidy story, and it fits the pattern. Fitting the pattern is not the same as being proven.
The case that it is not AI
The Stanford authors are unusually frank about the weaknesses of their own result, and anyone quoting the 19% should quote these too.
The gaps between more- and less-exposed young workers shrink when we account for education. Some differential trends are visible before the widespread use of generative AI. And the estimated gaps are larger in the ADP sample than in national survey benchmarks. In their words, this raises legitimate questions about how much of the pattern is actually caused by AI.
The timing objection is the strongest. One review of the evidence notes that some of the deterioration in AI-exposed occupations was underway in early 2022, months before ChatGPT was released. Commentators have pointed to other candidates from the same period, including interest-rate-driven hiring caution and over-hiring in tech during the pandemic. The paper’s authors say they tested and set aside several of these explanations, but the timing keeps the question open.
Other analyses reach a different conclusion. Some point out that hiring for junior and senior roles in exposed fields has fallen together since 2022, which looks more like a general slump than something aimed at juniors, and work using national administrative data elsewhere has questioned whether AI is the main driver of the early-career decline at all.
So the honest position is that the gap is real and its cause is disputed. It may be AI. It may be the interest-rate cycle landing hardest on the newest hires in the most cyclical white-collar sector. It is probably some of both, and nobody can yet say in what proportion.
What is not known
Where the missing juniors go. The data show fewer young people being hired into exposed occupations. They do not track whether those people end up in other careers, in further education, or unemployed.
What it does to the pipeline. Senior engineers are made from junior ones. If firms hire fewer juniors for a decade, the supply of experienced people eventually falls. Economists have started modelling this, but it has not yet happened, so there is nothing to measure.
Whether the pattern holds outside the United States. The Danish evidence suggests it may not, at least not yet.
Whether it persists when interest rates and hiring recover. A cyclical slump and a structural shift look identical until the cycle turns.
If you are early in your career
Any advice here is inference from a contested finding, not a tested strategy.
The exposed part of a junior role is the codified part: the work that can be specified fully in a ticket and checked against documentation. The protected part is everything that requires context. That argues for getting close to real systems and real users as fast as possible, through internships, open source maintenance, on-call rotations, anything that builds judgement no model has been trained on.
Using AI tools well is now table stakes, not an edge. The pattern in the data is that complementing uses of AI go with stable employment. Being the person who can direct, check and correct the tool puts you on that side of the line.
Do not read the 19% as a verdict on your prospects. It is a relative gap between groups, in one country’s payroll data, with a disputed cause.
If you hire
The pipeline point is yours to worry about. Cutting junior hiring because a model can do junior tasks saves money this year and may leave you with nobody to promote in five. Whatever the true cause of the current gap, a team that stops training people is making a bet about the future supply of seniors that no evidence yet supports.
Common questions
Is AI causing job losses?
Not at the level of the whole economy. Stanford’s 2026 analysis of US payroll data states that it does not see widespread, economy-wide job displacement associated with AI. The measurable effect is concentrated among the youngest workers in the most exposed occupations.
Are entry-level jobs disappearing because of AI?
Employment among 22 to 25 year olds in highly AI-exposed occupations is about 19% below where it would be had it kept pace with less-exposed peers, up from 15% a year earlier. The authors caution that part of the trend predates generative AI and shrinks when education is accounted for.
Why are junior workers affected more than senior ones?
The proposed explanation is that AI substitutes for codified, textbook knowledge, which is what new graduates offer, and complements the tacit knowledge that comes from experience. Experienced workers in the same occupations show no comparable gap.
Are young workers being laid off because of AI?
The data point to reduced hiring, not layoffs. The gap operates primarily through fewer young workers being taken on, and not through more of them losing jobs they already had.
The bottom line
The best available data say AI has not caused mass job losses and may be closing the entry door in the most exposed professions. The gap for young workers is real, growing, and only partly explained; interest rates and post-pandemic over-hiring have fingerprints on it too. If you are starting out, build the kind of judgement that is not in any training set. If you run a team, remember that the seniors you will need in 2032 are the juniors you hire now.
Did this answer your question?
Thanks. That helps us decide what to fix.
Leave a comment