Analysis · 31 July 2026 · 7 min read

Will AI replace entry level jobs?

Payroll data shows a 16 percent relative employment decline for workers aged 22 to 25 in AI exposed occupations. What it means if you are starting out.

Most claims about AI and jobs are forecasts. This one is not, which is why it deserves separate treatment.

Researchers at the Stanford Digital Economy Lab studied administrative payroll records from the largest payroll software provider in the United States and found something specific: workers aged 22 to 25 in the most AI exposed occupations saw a 16 percent relative decline in employment since generative AI came into wide use. Workers in less exposed occupations, and more experienced workers in the same occupations, held steady or continued to grow.

That is a measurement of what already happened, not a projection of what might. It is the sharpest single piece of evidence on this question so far, and it points at the part of the labour market that most coverage skips.

What the study found

The paper is Canaries in the Coal Mine by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen. Three things in it matter for anyone starting out.

The decline is concentrated by age, inside the same occupation. This is the finding that makes it credible. If exposed occupations were shrinking for ordinary economic reasons, you would expect everyone in them to feel it. Instead the effect sits on the youngest workers while their experienced colleagues in the same roles are stable. Something is happening to the entry rungs specifically.

It concentrates where AI automates rather than augments. The authors separate occupations where the technology substitutes for the work from those where it assists someone doing the work. The employment effect shows up in the first group.

The adjustment runs through employment rather than pay. Wages in these roles did not absorb the shock. Headcount did.

We should be equally clear about what the study does not establish. It measures an association between exposure and employment change in a period when a great deal else was happening in white collar hiring, including interest rates and post pandemic correction. The authors are careful about this and later work has probed the timing. Read it as strong evidence that something real is happening at the entry level, not as a clean measurement of AI causing a precise number of lost jobs.

Why entry level work is the exposed part

This lines up with what our own index measures, from a completely different direction.

Our scores come from 18 published sources rating how much of an occupation's task content overlaps with what current models can do, combined using the method set out on our methodology page. Across 961 occupations, the roles that score highest are the ones built on producing and processing text and structured documents.

Now think about what a first job in knowledge work actually consists of. The first draft. The research summary. The formatting pass. The routine ticket. The document review that a senior person checks afterwards. Entry level work in office jobs is, almost by definition, the teachable and checkable part of a profession, which is the same slice that current systems handle best.

So the two findings are describing one phenomenon from two angles. Our index says these tasks are the most exposed. The payroll data says the people who mostly do these tasks are the ones whose employment moved.

The occupations at the sharp end of this are not obscure. Computer Programmers (61) and Paralegals and Legal Assistants (50) both score above the middle of our index, and both are classic graduate entry routes into a profession. You can check any other role in the occupation index.

What this does not mean

It does not mean these careers are closed. It means the bottom rung is narrower, which is a different and more manageable problem.

A few things follow that are worth saying plainly, because the alarming version of this story is easier to write than the accurate one.

The occupation average is a floor on your exposure, not a description of it. If you are early in a text heavy field, your personal task mix is probably more exposed than the number attached to your job title, because you are doing the concentrated version of the exposed tasks.

Accountability is the durable part. In most professions the parts that survive are the ones someone has to sign for. Reaching those parts used to happen automatically by serving time on the draft work. If that path narrows, getting to judgment work faster becomes the thing to optimise deliberately rather than something that happens to you.

Exposure is not the same as disappearance. The mean score is 30. The median is 29. Even in our index, most occupations sit nowhere near the top, and a high score means a share of tasks are within reach of the tools, not that the profession ends. What happens next depends on adoption, cost, regulation and liability, none of which are in the score.

The evidence base is young. This is one study, on one country's payroll data, over a short window. It is the best evidence we have on this specific question and it is still one study. The same caution applies to our own index, and we keep an honest account of its limits rather than burying them.

If you are starting out now

The practical reading is not to avoid exposed fields. It is to be deliberate about which part of them you land in.

Prefer roles where you get near real decisions early, even at a smaller organisation, over roles where you will spend two years producing drafts that someone else signs. Treat fluency with the tools as table stakes rather than a differentiator, since the differentiator is being the person who can tell when the output is wrong. And when you compare two graduate offers with the same title, ask what the actual week looks like, because as our guide to reading your own job argues, the title is the least informative thing about it.

You can look up any occupation in the full index to see its score, the sources behind it and how much those sources disagree, which for early career decisions is often the most useful number on the page.

Figures come from the AI Job Risk Index and were current when this was published. Scores change as new research is ingested, so the index is always the live version. See how scoring works. Informational guidance based on published research, not professional career or financial advice.