Skip to content
NewsletterAI Learning, Keeping It HumanCall647-641-0643
Christopher Ross

Guides

The entry-level hiring gap is the clearest AI signal so far

By 6 min read

If you want the one place where the data and the AI-layoff headlines line up, look at workers aged 22 to 25. It is a hiring story, it is not yet shown to be caused by AI, and the people it affects mostly never appear in a layoff count.

Key takeaways

  • Stanford researchers report that employment among workers aged 22 to 25 in highly AI-exposed occupations is about 19 percent below the level it would otherwise have reached, relative to similarly aged workers in less-exposed jobs, as of June 2026. The figure was 15 percent in the July 2025 data. Experienced workers show no comparable gap.
  • The authors say the adjustment "appears to operate primarily through reduced hiring of young workers rather than increased separations," and they call their results "descriptive patterns, not causal estimates of the effect of AI."
  • The wider market is weak. September 2026 payrolls rose by 29,000 and the New York Fed puts recent-graduate unemployment at about 5.6 percent.

Most of the evidence in this series points one way: employers say AI far more often than the aggregate data supports. The one place where the data and the headlines meet is early-career work, and it is worth treating separately because it is a different kind of problem from a layoff. Nobody gets an announcement and nobody clears a desk. A person who was never hired does not appear in a count of people let go.

What Stanford found

Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at Stanford's Digital Economy Lab track employment by age and by how exposed an occupation is to AI. In their 12 August 2026 update, employment among workers aged 22 to 25 in highly AI-exposed occupations "now stands about 19% below" the level it would otherwise have reached, relative to similarly aged workers in less-exposed jobs. The shortfall was 15 percent in the July 2025 data. "Experienced workers show no comparable gap."

They say the adjustment "appears to operate primarily through reduced hiring of young workers rather than increased separations." The mechanism matters. It suggests employers are not removing the people they have so much as declining to add the next person in certain roles.

What the authors say they cannot claim

The authors are direct about limits. They describe the results as "descriptive patterns, not causal estimates of the effect of AI," and on whether AI is causing the changes they write, "We cannot yet answer that question definitively." They note that the gaps shrink when education is controlled for, that some differential trends predate widespread generative AI use, and that the estimates are sensitive to specification choices, meaning the number moves depending on how the analysis is set up. They conclude that no single study, theirs included, is definitive.

A February 2026 note from the same group tested interest rates as an alternative explanation. It found no strong support for it, since more AI-exposed occupations tend to be less exposed to rates. With controls for each firm in each period, the decline for the most exposed young workers becomes statistically significant only from 2024, not 2022 or 2023. Of the earlier declines, the authors say they are "likely (at least partly) due to some combination of other factors, not just AI."

The market around it

The wider picture is soft. The Bureau of Labor Statistics reported that nonfarm payrolls rose by 29,000 in September 2026, with unemployment at 4.2 percent. It revised July to a loss of 10,000 jobs and August to 133,000 jobs added (BLS, 2 October 2026). The New York Fed says unemployment for recent college graduates aged 22 to 27 stayed elevated at about 5.6 percent in the second quarter, with underemployment at 42 percent (New York Fed). The page I read gives no all-worker rate or breakdown by major, so I am not quoting comparisons from it.

In a labour market that adds so few jobs, employers can hire fewer new graduates without needing a reason. Oxford Economics, in its January briefing, called the rise in graduate unemployment "cyclical rather than structural" (as reported by Fortune). Stanford's data and Oxford's reading are not obviously in conflict. One is about a specific, exposed group of occupations and the other is about graduates overall.

What I take from it

This is my reading and not a finding. The most defensible claim the evidence supports is that AI may be changing who gets hired into certain entry-level roles, and that the layoff headlines are the wrong place to look for it. Firms may be using AI tools to cover work they used to staff with new graduates, or freezing those roles while they spend elsewhere, or the pattern may be cyclical. The Stanford authors say they cannot tell which.

For someone early in a career, the practical response does not depend on settling the causal question. Roles that start with routine, checkable tasks are the ones a tool can cover first. Finished work you can show, and the habit of checking an AI's output carefully, are harder for a hiring manager to wave off. For a manager, the useful question is what the junior role was for. If it was where people learned to do the senior work, removing it has a cost that will not show up for a few years.

That second point is where this series meets the learning questions I care about most. I would watch the experienced-worker numbers on Stanford's monthly dashboard. If that gap opens, I will change my reading and say so. The broader case is in AI is in the budget more than in the headcount, and the earlier explanation executives gave in 2022 and 2023 is in the pandemic hiring piece.

How this was made. AI tools did the source reading and checking for this piece, and AI reviewers argued against the draft before it went out. The question is mine, and so is the decision to publish it.

Common questions

Is AI taking entry-level jobs? There is a measurable gap: Stanford researchers find employment for 22 to 25 year olds in highly AI-exposed occupations about 19 percent below the level it would otherwise have reached, relative to similarly aged workers in less-exposed jobs, as of June 2026. They describe it as descriptive and not causal, and say it works mainly through reduced hiring and not layoffs.

Could something other than AI explain it? Possibly. A weak overall labour market, with 29,000 payroll jobs added in September 2026, lowers hiring of new workers. The Stanford authors report that interest rates do not explain the pattern well, but they also note that some trends predate widespread generative AI.

References

Checked against the sources above, as listed, on 2026-10-29.