Stanford study links AI adoption to widening entry-level employment gap

New research from Stanford University economists indicates that artificial intelligence is causing significant job losses for younger workers in specific sectors, even as older employees remain largely unaffected. The updated study suggests that the impact of AI on the labour market is not uniform, with entry-level roles in high-exposure occupations experiencing a marked decline in employment levels. This finding challenges previous assumptions that AI would disrupt the workforce broadly and instead points to a specific vulnerability among those at the start of their careers.

The August 2026 edition of the paper, titled Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, revises a study published the previous year with fresh data and refined statistics. The researchers found that the employment trends identified last year are persisting and expanding. Specifically, employment levels for workers aged 22 to 25 in the most AI-exposed occupations are now 19 per cent below those of their peers in fields less exposed to AI disruption. This gap has widened significantly, as the difference measured just 13 per cent in the prior year’s analysis.

To derive these figures, the research team utilised a large subsample of anonymised, high-frequency payroll data aggregated by HR management company ADP. They assessed each occupation’s exposure to AI disruption using two primary metrics. The first was a potential labour market impact gauge established by previous researchers. The second was the Anthropic Economic Index, which analyses how various occupations actually use the Claude model in everyday work. A similar report based on occupational Gemini usage was released by Google last month, providing further context to these usage patterns.

When analysing the data across the entire economy, the researchers found little to no difference in relative overall employment between jobs judged most and least affected by AI. However, when isolating workers aged 22 to 25, a distinct pattern emerged. Since 2022, employment in the top 40 per cent of AI-impacted jobs had fallen by approximately 11 per cent. In contrast, total employment for these young workers in the 60 per cent of jobs with the least AI impact grew by 10 per cent over the same period.

Further analysis revealed that this phenomenon is primarily driven by lower hiring rates for entry-level workers in AI-impacted fields, rather than increased firings or voluntary resignations. The labour market effects for this age group were mostly seen in reduced overall employment numbers, with little evidence of reduced pay rates. The researchers noted that not all jobs with potential for AI disruption are equal. Using Anthropic’s Economic Index, they differentiated between automative tasks, which fully replace human work, and augmentative tasks, which help human workers become more effective. Occupations such as accountants, auditors, and receptionists were judged most susceptible to automation, while roles like chief executive and registered nurse showed high levels of AI augmentation.

The data indicates that jobs where AI automation is prevalent are showing the worst relative employment levels for entry-level workers. The researchers concluded that automation-oriented uses of AI are substituting for labour, while complementary uses are associated with flat or rising employment. They theorised that entry-level workers are particularly affected by AI’s impact on jobs requiring heavily codified knowledge, such as formal, standardised, and documented information that can be taught through education or written procedures. This contrasts with jobs where AI complements the tacit knowledge of experienced workers, which is acquired through practice, mentorship, and repeated exposure to real situations.

To test this hypothesis, the researchers used the required level of formal education in the O*NET occupational database as a proxy for reliance on codified knowledge. They found that occupations with higher codified knowledge experienced slower entry-level employment growth, while those with higher tacit knowledge saw faster employment growth for mid-career and senior workers. Additionally, higher education appears to serve as a buffer against these effects. Occupations with a higher share of college graduates showed more muted differences between more-exposed and less-exposed roles. In jobs with few college graduates, the least AI-exposed occupations saw job growth, while the most exposed occupations experienced declining employment.

Lead researcher Erik Brynjolfsson warned in a recent interview with The Washington Post that current trends suggest a future where jobs for those employed in the pre-AI era largely persist, while many jobs for the incoming working-age cohort begin to disappear. He stated that the entry-level effects being measured are real, persistent, and widening. Brynjolfsson expressed concern about a labour market that maintains its overall employment level while quietly closing the on-ramp for people starting their careers.