AI's Impact on the Labor Market: New Metrics and Early Evidence

A new metric called observed exposure measures how much AI has actually entered the real labor market. It combines O*NET tasks, usage data from the Anthropic Economic Index, and β estimates that look at whether LLMs can make tasks 2x faster.

This metric reflects automation usage in actual work contexts, work-related usage, and AI's share within a job as a whole, more than mere theoretical possibility. As a result, while AI can do a lot, its actual coverage is still much smaller than the theoretical ceiling.

The core findings are clear. The Computer & Math occupational group is broadly exposed in theory, but actual coverage is around 33%, and the occupations with the highest exposure are Computer Programmers (75%), Customer Service Representatives, and Data Entry Keyers (67%). Conversely, 30% of workers showed effectively zero exposure in the sample, including occupations such as cooks, motorcycle mechanics, rescue workers, bartenders, and dishwashers.

The more exposed an occupation was, the weaker its BLS 2024–2034 employment growth projection tended to be. For every 10-percentage-point increase in exposure, the growth projection fell by 0.6 percentage points, though this relationship was not strong, and the same correlation did not appear when using the theoretical β metric alone.

Worker characteristics also differ. The highest-exposure group is more female, more highly educated, and has average wages that are 47% higher. Based on the CPS from August–October 2022, the share holding a graduate degree diverged sharply: 4.5% in the non-exposed group versus 17.4% in the high-exposure group.

Early employment indicators do not yet show a large-scale shock. There is no evidence that unemployment among highly exposed workers has systematically risen since late 2022, but the data does suggest a possible slowdown in hiring of young workers in high-exposure occupations.