Source: Labor Market Impacts of AI: A New Measure and Early Evidence

Summary: Anthropic researchers introduce "observed exposure," a measure that blends theoretical AI task capability (from Eloundou et al.) with actual Claude usage data, weighting automated and work-related uses more heavily. Applied to ~800 US occupations, they find AI covers far less than what's theoretically possible — e.g., only 33% of Computer & Math tasks despite 94% theoretical feasibility. Comparing high-exposure workers to low-exposure workers since ChatGPT's release, they find no meaningful rise in unemployment, though hiring of young workers (22–25) into exposed occupations shows a tentative ~14% slowdown.

Key Takeaways:

• Computer programmers (75%), customer service reps, and data entry keyers are the most AI-exposed occupations; 30% of workers have zero exposure
• Workers in the most exposed jobs tend to be older, female, more educated, and higher-paid — earning 47% more on average
• BLS projections correlate slightly with observed exposure: each 10pp increase in coverage maps to 0.6pp less projected job growth through 2034
• No statistically significant increase in unemployment for highly exposed workers since late 2022
• Suggestive evidence that hiring of 22–25 year-olds has slowed in exposed occupations, echoing findings from Brynjolfsson et al.

Written by Pi, using my tldr skill and Opus 4.6