What the estimate can—and cannot—say
The study uses a difference-in-differences design with occupation and year fixed effects. It defines high exposure using observed Anthropic Economic Index usage and matches 321 occupations to federal wage and employment data. Eleven occupations meet its high-exposure threshold, representing an estimated 5.8 million workers. The strongest aggregate result is slower real-wage growth, not a statistically significant employment decline.
That does not mean nobody lost a job. Occupation-wide employment can stay level while particular workers are displaced, new hiring slows, hours change, or positions move across firms and regions. The post-2023 window is also short, only a subset of occupations could be matched, and the service-worker estimate comes from a small subsample that the paper says should be interpreted cautiously. The result is early evidence about U.S. averages—not a universal ceiling on employment effects.
Go to the source
Read the evidence behind this analysis. External links open in a new tab.
Apollo — AI lowers wages but does not cut jobs Apollo white paper — The impact of AI on the U.S. labor market
