Anthropic Economic Index report: Learning curves

Source: Anthropic Research Date: 2026-03-24 URL: https://www.anthropic.com/research/economic-index-march-2026-report

Summary

Economic Index March 2026: ~1M conversations from Feb 5-12, 2026. Users with 6+ months tenure show 10% higher success rate, shift 7pp more usage toward work tasks, and exhibit ~one additional year of educational complexity per year of tenure. Experienced users concentrate Opus on higher-wage occupational tasks, with effect roughly 2x larger for API developers vs. web users.

Implications

This is the AI skill formation + inequality thread getting empirical teeth. Learning-by-doing in AI tool use is real and measurable — experienced users succeed more and deploy more capable models more selectively. The “one additional year of educational complexity per year of tenure” is a striking compound — it implies early adopters are pulling away from newcomers in terms of what they can accomplish. This is the mechanism behind the geographic concentration story: high-GDP regions that adopted early are compounding advantages. Watch whether this shows up in hiring data — early-adopter employees may be commanding AI-related premiums that aren’t yet visible in aggregate compensation data.