AI's Real Job Impact: Smaller Than the Forecasts, Harder on New Grads

Dario Amodei has predicted that AI could wipe out half of all white-collar jobs and push US unemployment to 20 percent. A new policy brief from the Stanford Institute for Economic Policy Research (SIEPR) now tests that forecast against the actual labor data, and the answer is more complicated than either the alarm or the dismissal.

Unemployment for the most AI-exposed workers has risen 0.77 percentage points since 2022. For workers in the least AI-exposed occupations, it has risen 0.85 points. The least-exposed group has risen slightly more, a small inversion of what a strict "AI is eating jobs" story would predict, though both movements are small enough to read as a broadly softening labor market rather than an AI-specific shock. That empirical center is the SIEPR policy brief's main finding on the aggregate.

The same team's working paper, "Canaries in the Coal Mine", built on ADP payroll data through September 2025, finds a 13 percent relative decline in employment for early-career workers aged 22 to 25 in the most AI-exposed occupations, even after controlling for firm-level shocks. Software developers in that age band are down roughly 20 percent since late 2022 in the same dataset. The new-graduate labor market is a specific pocket where AI may be a real factor alongside the broader softening, and it is too sharp to wave away as "the economy."

Worker productivity, the third meter, splits engineers in a Hacker News thread on the SIEPR brief into two camps: those who say AI acts as a Pareto-sharpener, widening the gap between the top 20 percent and the median, and those who say AI compresses the field, hurting top performers while lifting juniors. The data has not picked a winner. Controlled studies generally show positive effects, and the brief itself flags firm AI adoption as real but uneven.

Aggregate unemployment is up slightly, with the least-AI-exposed group rising marginally more. The 22-to-25 cohort in AI-exposed occupations is down 13 percent. Productivity is contested. The three meters point in different directions, and the SIEPR synthesis asks readers to hold all of them at once.

The Amodei forecast, the 20-percent-unemployment claim, is reading one of the three meters. The dismissals are reading another. The policy brief is asking readers to hold all three at once. When the next AI-jobs headline lands, the question worth asking is which workers, which time horizon, and which measure, because the answer depends on which of those you are looking at.

The SIEPR working paper cuts off at September 2025. The next reading of the three meters lands with whatever ADP and BLS publish through the rest of the year.