Apollo discovers that AI is affecting individual paychecks instead of overall payrolls.

      An Apollo analysis of 321 occupations found that wages in jobs significantly impacted by AI increased 6.7% more slowly after 2023, with no statistically notable effect on employment. The disparity was 10.7% in the lowest-paid quartile and none in the highest.

      The first noticeable influence of AI on the labor market is not unemployment. According to Apollo’s chief economist Torsten Slok, the employment effect so far has been negligible, while the evident issue is with wages. His collaboration with Sania Edlich revealed that wages in highly AI-exposed occupations grew 6.7% more slowly after 2023 in comparison to those in low-exposure jobs, and employment in those roles exhibited no statistically significant change.

      The distribution of these effects is particularly noteworthy. The wage gap was 10.7% in the lowest wage quartile, 5.4% in the second quartile, and 4.0% in the third quartile, with no significant impact in the top quartile.

      The methodology employed is unique and deserves mention. The authors aligned 321 occupations with labor statistics from 2015 to 2025, utilizing the Anthropic Economic Index, which gauges actual AI usage derived from real model interactions rather than theoretical exposure.

      They are transparent about the limitations of their study. Exposure is derived from usage data from only one company, only 321 of approximately 800 occupations were matched, and their most striking statistic—a 24.3% gap for service workers—comes from a small subsample that should be interpreted cautiously.

      Contrary evidence suggests otherwise. US statisticians identified a 0.2% decline in jobs across 18 occupations heavily affected by AI, while overall payrolls increased by 0.8%. Goldman Sachs reported sharper declines in job openings in areas prone to AI substitution, within a market where new entrants are already facing challenges.

      Diane Gherson, former chief human resources officer at IBM, proposes that job losses may not be immediately visible. Companies are discreetly reducing hires for high-turnover, lower-wage positions instead of publicly announcing layoffs.

      She also highlights an accounting distortion that favors this trend. Severance packages can be recorded as one-time restructuring costs that investors overlook, whereas retraining expenses appear in operating costs each quarter, making layoffs seem more favorable on financial statements than reskilling efforts.

      A contrasting example she references is European: Ikea transformed call center employees into remote interior design advisors after automating much of their previous tasks, resulting in a service reportedly valued at around €1.3 billion.

      Slok’s broader assertion is that the economy is becoming more dynamic rather than contracting, with business formation at its highest recorded rate. However, he acknowledges that the benefits in productivity remain unproven since profit margins outside of the largest tech companies have not yet improved.

      Notably, Europe lacks a similar study, which is a critical gap. The wage effects that Apollo highlights would not be apparent in most European labor data, and TNW has already reported on the impacts of AI on jobs in the region without measuring wages in this fashion.