Is AI closing the door on entry-level jobs?
When people ask whether AI will take their job, they usually picture layoffs. The most careful data so far points somewhere else: not at the people already inside the building, but at the ones trying to get in.
The evidence comes from a study by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at Stanford’s Digital Economy Lab. They analysed payroll records from ADP covering millions of US workers through June 2026. The paper is called “Canaries in the Coal Mine?”, after the birds miners once used as an early warning. It is a working paper, not yet peer-reviewed, and its authors describe the results as early and descriptive.
The short version: there is no wave of AI layoffs. But employment of 22- to 25-year-olds in the most AI-exposed jobs is about 19% lower than it would otherwise be, mostly because fewer of them are being hired. Whether AI is the cause is not yet proven.
What this means for you
If you manage a team, the risk isn’t only what AI does to today’s roles. It is what happens to the pipeline. Junior roles are where people build the tacit knowledge that makes them valuable later. If those roles thin out, someone has to decide how your organisation grows its next generation of experienced people.
If you are earlier in your career, the finding points in a useful direction. The work that holds up is the work that depends on experience and judgment. That suggests seeking out situations where you learn by doing, and by working alongside people who have done it before, rather than only producing documented, standard output.
And for everyone, the question from our last post applies here too: which parts of your work are codified and easy to reproduce, and which depend on what you have learned by doing? That is a question worth answering deliberately, before the market answers it for you.
What the study found
The headline is reassuring: the authors find no evidence of widespread, economy-wide job displacement from AI.
The detail is not. They sorted occupations by how exposed they are to AI, meaning how many of a job’s tasks a language model could do. Among workers aged 22 to 25 in the most exposed occupations, employment is now about 19% lower than it would be had it kept pace with young workers in less-exposed jobs. A year earlier, the same measure showed 15%. In plain numbers, employment for this age group in the most exposed jobs fell about 11% between November 2022 and June 2026, while it grew about 10% in the least exposed ones.
Experienced workers in the same occupations show no comparable gap.
Not layoffs, but hiring that never happens
The gap mostly comes from fewer young people being hired, not from more of them being let go. That is why it is easy to miss. Nobody gets an email announcing that a role no longer exists. The job simply doesn’t get posted.
Pay tells a similar story. So far, the adjustment shows up in employment rather than in base salaries.
Where AI replaces and where it assists
The declines are concentrated in occupations where people mostly use AI to automate tasks. Where AI mainly assists the person doing the work, employment is flat or rising, especially for experienced workers. (The study draws this distinction from how people actually use Anthropic’s Claude, among other measures.)
The revised paper adds a second pattern: codified versus tacit knowledge. Young workers lose ground in occupations built on written-down, teachable knowledge: procedures, textbooks, standard methods. Experienced workers gain in occupations that rely on tacit knowledge, the kind you build through practice, mentoring and repeated exposure to real situations. The authors call this suggestive, not proven, but it fits what AI is good at: applying what has already been written down.
It also echoes the finding of the Princeton study we wrote about recently: AI handles routine, well-documented work better than open-ended judgment.
What we can’t say yet
The authors are unusually direct that this is not proof AI is the cause. Their own caveats:
- The gap shrinks when they account for education levels. That could mean education is an alternative explanation, or that it is the very channel through which AI acts.
- Some exposed and less-exposed occupations were already diverging before ChatGPT arrived.
- The gap is larger in the ADP data than in national surveys.
On the other side, interest rates, remote work and the tech sector don’t explain the pattern. The gap kept widening long after interest rates peaked. A separate US Census Bureau working paper using government data reports a similar direction, with early-career hiring falling in the most AI-exposed industries.
The honest summary: a real pattern, a plausible link to AI, and no final answer. The authors say themselves that no single study settles it, and they now publish a monthly-updated dashboard so readers can follow the numbers as they change.
Working out which parts of your work are codified and which are tacit is exactly what we do in coaching. If you’d like to explore it for your role or your team, let’s talk.
Sources are linked inline. The Stanford study is a working paper that has not been peer reviewed, and its results are descriptive rather than causal.