AI Job Losses Hit Entry-Level Workers, Stanford Finds

New research from Stanford University economists suggests that the early impact of artificial intelligence is showing up most clearly in the first rung of the career ladder. Employment for workers aged 22 to 25 in occupations highly exposed to AI is now 19% below employment in less exposed fields, according to the August 2026 update covered by Ars Technica.

The Entry-Level AI Employment Gap Is Widening

The researchers compared anonymized, high-frequency payroll data aggregated by ADP with measures of occupational AI exposure. The gap has widened from 13% last year to 19% in the latest analysis. Since 2022, employment for 22-to-25-year-olds in the 40% of occupations most affected by AI fell about 11%, while employment in the 60% of less-affected occupations grew 10%.

The pattern does not appear to be a broad economy-wide collapse. Across workers of all ages, the study found little relative employment difference between highly exposed and less exposed occupations. The sharper divide is concentrated among younger workers trying to enter the labor market.

Automation Hurts More Than Augmentation

The study draws an important distinction between AI that automates work and AI that helps people do their jobs better. Accountants, auditors, receptionists, and information clerks are among the occupations more associated with automative use, while roles such as executives and registered nurses show more augmentative use in the Anthropic Economic Index cited by the researchers.

That distinction maps onto the employment data. The researchers write that automation-oriented uses are consistent with labor substitution, while complementary uses are associated with flat or rising employment. In other words, the question is not simply whether a job touches AI, but whether AI is deployed to remove a task or expand a worker's capability.

What Builders Should Take From the Data

For AI builders, the findings are a warning against measuring success only by task completion. A system that replaces the junior analyst who previously learned through routine work may deliver short-term efficiency while weakening the talent pipeline that produces experienced operators.

Product teams should therefore design for augmentation where possible: preserve review steps, expose reasoning and sources, create feedback loops, and give early-career users structured opportunities to build judgment. This is the same design tension now visible in physical AI startups such as General Intuition, where systems are being built to act in the world rather than merely answer questions.

Key takeaway: AI is not affecting every worker equally. The current evidence points to a concentrated entry-level employment shock in occupations where companies use AI primarily to automate work, making augmentation a product and workforce-design choice rather than a slogan.