The data they chose to analyze shows employment for workers ages 22 to 25 falls most in highly-exposed occupations

For years, observers have argued that highly capable AI systems could eventually perform most human work at lower cost, raising the prospect of a labor-market shock.

Now, fresh work from Stanford University economists point to a narrower but already measurable effect: sizable entry-level losses for younger staff in some occupations, while older workers look largely untouched so far.

The authors say last year’s pattern for entry-level workers (mostly in the US) is still visible and is spreading. Employment among people ages 22 to 25 in the occupations judged most exposed to AI now sits 19% below the level for peers in jobs less exposed to AI. That gap was 13% for the preceding year.

How they analyzed the data

To produce those figures, the team drew on a large anonymized, high-frequency payroll sample from HR firm ADP. Occupations were scored for AI exposure with a potential labor-market impact measure from earlier research examined that line of work critically earlier this year) and with the Anthropic Economic Index, which tracks how occupations actually use Claude.

Across the whole economy, the data showed little or no difference in relative employment between jobs ranked most and least AI-affected on those metrics. Restricting the sample to ages 22 to 25 changes the picture:

  • Since 2022, employment in the top 40% of AI-impacted jobs has dropped by about 11% that age group.
  • In the 60% jobs with the least AI impact, employment for the same young workers rose 10% over the same span.
  • Economy-wide, relative employment effects in more heavily AI-impacted fields remain relatively muted, according to Brynjolfsson et al. For entry-level workers, the gap between fields that are and are not AI-impacted is large and still widening, the same paper shows.
  • A closer look indicates the pattern is driven mainly by weaker hiring of entry-level workers in AI-impacted fields, not by more firings or quits. Among this age group, the labor-market effect shows up mostly as lower overall employment rather than lower pay.
  • Not every job that looks open to AI disruption is the same. Anthropic’s Economic Index splits queries into “automative” uses that fully replace work a person used to do and augmentative uses that help people do work they are still needed for. On that split, accountants and auditors and receptionists and information clerks rank among the roles most open to automation, while chief executives and registered nurses rank among those that use AI most often as an aid.
  • Jobs where automation is common now show the weakest relative employment for entry-level workers.

The researchers write: “The findings are consistent with automation-oriented uses of AI substituting for labor while complementary uses are associated with flat or rising employment.” They also suggest entry-level staff may be hit hardest in jobs that rest on heavily “codified” knowledge: formal, standardized, documented material that can be taught in school, textbooks, or written procedures. That stands in contrast to work where AI mainly complements an experienced person’s “tacit” knowledge, which they say is overwhelmingly gained through practice, mentorship, and repeated real situations.


Easier to automate: easily codified skills

The researchers tested the idea by treating required formal education in O*NET’s occupational database as a stand-in for reliance on codified knowledge:

  • Occupations with more codified knowledge show slower entry-level employment growth, while those with more tacit knowledge show faster growth for mid-career and senior workers.
  • Higher education may still blunt the effect. Occupations with a larger share of college graduates showed more muted differences between more-exposed and less-exposed jobs.
  • In occupations with few college graduates, the least AI-exposed jobs grew while the most exposed declined.

The lead researcher of the study has cautioned that today’s pattern could point towards a near future in which pre-AI jobs largely remain while many openings for the next working-age cohort fade. This could reflect a labor market that keeps its overall employment level while quietly closing the on-ramp for people starting out in their careers.