The Age Gap in AI Displacement
AI's workforce impact isn't landing evenly. In the United States, the clearest signal so far is generational: early-career workers in AI-exposed occupations are losing jobs, while workers in the same occupations aged 26 and older keep growing. Employment for software developers aged 22–25 has fallen close to 20% from its 2022 peak as of Sept 2025.
Key finding: Among workers aged 22–25, the most AI-exposed occupations show a 16% relative decline in employment compared to the least-exposed occupations — after controlling for firm-level hiring swings. The same gap does not appear for workers 26 and older. This is not a recession pattern; it's a seniority-biased technological shift.
Employment change, late 2022 → September 2025
% changeEarly-career software developers have lost a fifth of their headcount since 2022. For workers aged 26 and older in the same occupations — and for workers of any age in low-AI-exposure occupations — employment has continued to grow.
Why this matters for AI job risk
Our Capability Coverage Index scores which tasks AI can perform today. The Brynjolfsson et al. findings show where that capability is already showing up in the labor market: entry-level roles in exposed occupations. The mechanism is intuitive — junior workers perform more of the tasks AI is best at today (writing boilerplate code, resolving routine support tickets, formatting documents), and their labor is the most readily substitutable. Senior workers in the same occupations hold judgment-heavy work that AI still can't reliably do, and their headcount keeps rising.
The practical implication for anyone early in a career in an exposed occupation: the job title is less important than the task mix inside it. A junior software developer writing CRUD endpoints faces different displacement pressure than a junior software developer debugging distributed systems. A junior support agent handling tier-1 tickets faces different pressure than one escalating edge cases. Your occupation's headline risk score is the starting point; the tasks you personally own are what determine your risk.
The authors call this the “canary in the coal mine” pattern — early-career workers are the first signal of where AI is binding on labor. If the pattern extends to mid-career cohorts in coming years, the 2022–2025 decline in entry-level hiring is a leading indicator for broader displacement.
Sources
- Brynjolfsson, Chandar, and Chen (2025), “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence.” Uses anonymized ADP payroll data covering approximately 62 million U.S. workers across 285,000 firms, linked to occupational AI exposure measures.
- Key figures: −20% decline for software developers aged 22–25 vs. 2022 peak; −15 log-point relative decline for most-exposed vs. least-exposed quintile ages 22–25 (paper rounds this to 16%); +5% to +13% growth in the lowest three exposure quintiles across all age groups; −6% change in high-exposure quintiles for 22–25 year olds over the same window.
- AI exposure measured using the Eloundou et al. (2024) GPT-4 occupational task exposure scores, with Anthropic's Claude-usage-based exposure measure as a robustness check. Both measures produce qualitatively similar results.
- Our Capability Coverage Index uses a separate task-level exposure measure grounded in O*NET tasks and Anthropic's observed AI usage patterns. See our full methodology for details.