Generative AI linked to 16% drop in employment for 22–25‑year‑olds in exposed occupations

A Stanford Digital Economy Lab working paper released in November 2025 reports a 16% relative decline in employment for workers aged 22 — 25 in occupations most exposed to generative AI after the technology spread, signaling a measurable reduction in entry‑level hiring that could weaken the pipeline for early‑career talent. This drop matters because those initial jobs historically provide the training and experience that build long‑term professional skills. An Anthropic report from March 2026 produced suggestive evidence pointing in the same direction, reinforcing the Stanford finding rather than overturning it. Both pieces of empirical work, while not definitive on causation, add weight to a pattern observed across multiple data sources.

The decline appears concentrated in roles where generative AI is used heavily. Occupations cited include software developers, computer programmers, information‑systems managers and customer‑service representatives. The Stanford paper notes that more experienced workers in those same occupations did not see the same employment decline, and that entry‑level roles with low AI exposure likewise did not fall, implying that AI is substituting for junior tasks rather than broadly eliminating positions across experience levels.

Those occupation‑specific shifts occur against broader weakness for recent graduates. The Federal Reserve Bank of New York reported that in Q4 2025 the unemployment rate for recent college graduates rose to 5.6%, while underemployment reached 42.5%, its highest level since the COVID surge. Analysts caution that muted hiring after the pandemic complicates attribution, but the timing and concentration by occupation make generative AI a plausible accelerator of these trends.

The implications extend beyond short‑term job counts because entry‑level roles serve as on‑the‑job training. If firms use AI to absorb drafting, triage, coding, summarizing and administrative preparation that previously taught judgment, systems debugging and client interaction, short‑term efficiency gains could produce long‑term skill erosion. Early‑career workers already report longer searches, greater financial precarity, anxiety and burnout; lost training opportunities risk delayed independence and weaker professional foundations.

Observers and the papers’ authors argue that a coordinated response is needed now. Proposed measures include reorienting educational curricula for AI‑augmented workplaces, government policies that incentivize businesses to hire and train early‑career staff, and corporate investment in apprenticeship‑style development that pairs AI tools with human mentoring. Experts also recommend students pursue applied AI fluency — learning how to use models in context rather than relying solely on coding skills — to help preserve training pathways while adapting to changing job tasks.