Stanford Study Finds AI Is Hitting Entry-Level Jobs Hardest

A new study from the Stanford Digital Economy Lab, published on August 24, 2026, finds that artificial intelligence is landing its sharpest blow on the youngest workers entering the labor market. Titled «Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence», the paper updates a version the same team published a year earlier. Its lead author is Erik Brynjolfsson, the lab's director, working with Bharat Chandar and Ruyu Chen.

The central finding is stark: employment among workers aged 22 to 25 in occupations considered most exposed to AI is now roughly 19% below employment in occupations less exposed to the technology. A year ago, when the same team last measured it, that gap stood at 13%. The gap, in other words, is not just present — it is widening.

To reach that conclusion, the researchers drew on a large subsample of anonymized, high-frequency payroll data aggregated by ADP, the largest payroll processor in the United States. The dataset covers millions of workers across more than 730 occupations — roughly one in six American workers. The team measured each occupation's exposure to AI using two separate tools: an established labor-market impact indicator from earlier research, and the Anthropic Economic Index, which tracks how each occupation actually uses the Claude model day to day. A month earlier, Google had published a similar report based on real-world Gemini usage.

No mass layoffs — but a hiring tap that is closing

Across the economy as a whole, the researchers found little to no difference in relative employment between the most and least AI-exposed occupations. The effect is concentrated almost entirely among the youngest workers. Since 2022, employment among 22-to-25-year-olds in the 40% of occupations most affected by AI has fallen by about 11%, while employment among young workers in the 60% of occupations least affected has risen by about 10% over the same period.

TIME, reporting on the same study, cites a separate measure: in occupations most exposed to AI automation — software engineering, marketing and customer service among them — entry-level workers saw a 16% relative decline in employment, even after controlling for effects specific to individual companies. Ars Technica and TIME frame these as distinct angles on the same dataset, not a single number restated twice, and the underlying pattern holds across both: the damage lands on hiring, not on people already in a job. The researchers found the effect works mainly through a drop in hiring of entry-level workers in affected occupations, not through more layoffs or resignations. It shows up mostly in the overall employment level, not in wages.

• 19%: the employment gap for 22-to-25-year-olds in the most AI-exposed occupations versus their peers, up from 13% a year earlier
• 11% vs. 10%: since 2022, employment for young workers has fallen about 11% in the 40% of occupations most affected by AI, while rising about 10% in the 60% least affected
• 16%: the relative decline in entry-level employment measured separately in the most automation-exposed occupations (software engineering, marketing, customer service), controlling for firm effects (TIME)
• 6% to 12%: employment growth for workers 30 and older in high-AI-exposure occupations between late 2022 and May 2025 (TIME)

Why some jobs and not others: automation versus augmentation

The Anthropic Economic Index draws a distinction between «automative» uses of AI, where the technology fully replaces a task a person used to do, and «augmentative» uses, where it helps a worker do their job more efficiently. Occupations like accountants and auditors, and receptionists and information clerks, rank among those most likely to be automated. Occupations like chief executives and registered nurses tend to use AI in a more augmentative way.

Occupations where automation-oriented use of AI dominates show the worst relative employment outcomes for entry-level workers. As the researchers put it, in a line reported by Ars Technica:

The findings are consistent with automation-oriented uses of AI substituting for labor while complementary uses are associated with flat or rising employment.

The researchers' working hypothesis is that entry-level workers are hit hardest in occupations that lean heavily on «codified» knowledge — formal, standardized knowledge that can be taught through schooling, manuals and written procedures — as opposed to occupations that rely on «tacit» knowledge, acquired through hands-on practice, mentorship and repeated experience.

Using the level of education required for an occupation, drawn from the O*NET database, as a proxy for how much it depends on codified knowledge, the researchers found that highly codified occupations show slower entry-level employment growth, while occupations built on tacit knowledge show faster employment growth for mid-career and late-career workers.

Higher education appears to cushion the effect. In occupations with a high share of college graduates, the difference between highly exposed and less exposed jobs is more muted. In occupations with few college graduates, the least-exposed jobs saw employment grow while the most-exposed jobs saw employment decline.

TIME reports a further data point: among workers 30 and older in high-AI-exposure categories, employment grew between 6% and 12% from late 2022 to May 2025 — a possible sign that older workers hold tacit knowledge that is harder for AI to replicate, or organizational standing that offers more protection. More manual occupations — home health aides, maintenance workers, taxi drivers — saw employment hold steady or even grow, according to TIME.

Our findings are consistent with the hypothesis that AI is having this effect [on the labor market], especially for entry-level workers.

Erik Brynjolfsson, the lab's director, described the pattern as unusually sharp when speaking to TIME:

It was really striking to see such a sharp effect for certain categories and not others. I think it's fair to say that technology has always been destroying jobs and always been creating jobs. If we want to create not just higher productivity, but widely shared prosperity, using AI to augment and not just automate work is a good direction to go.

The entry-level effects we're measuring are real, persistent and widening, and I'm more worried than I was about a labor market that keeps its overall employment level while quietly closing the on-ramp for people starting their careers.

The irony is not lost on the researchers themselves: they used AI to write the code that processes the ADP data, to check references, to generate and edit the study's charts, and to help with the writing — a use they describe, per TIME, as squarely «augmentative».

What this changes for young workers and employers

For young graduates and the companies weighing whether to automate or augment a role, the Stanford data is a caution rather than a verdict: the effect runs mainly through fewer entry-level openings rather than existing staff losing their jobs. Employers who use AI to make junior staff more capable, rather than to remove the need for them altogether, appear — on this evidence — to be the ones still hiring.