Anthropic launches the Anthropic Economic Index

Anthropic launched the Anthropic Economic Index, a recurring research initiative that mines anonymised Claude.ai conversations for evidence of how AI is actually being used across the economy, rather than how it is expected to be used. The first report analysed roughly one million conversations, automatically classified against the roughly 20,000 work tasks catalogued in the US Department of Labor’s O*NET database, and Anthropic open-sourced the resulting dataset for other researchers to build on.

The headline finding was concentration: 37.2% of conversations fell in the “computer and mathematical” occupational category, a share far above that category’s share of the US workforce, followed by arts, design and media work at around 10%. Looking within occupations rather than across them, Anthropic found usage was shallow more often than deep — about 36% of occupations used Claude for at least a quarter of their listed tasks, but only around 4% used it for three-quarters or more. Anthropic also split usage between “augmentation,” where a human retains control of a task, and “automation,” where the model completes it directly, and reported a roughly 57/43 split toward augmentation. Adoption skewed toward mid-to-high-wage roles such as programmers and data scientists, with both the lowest- and highest-paid occupations showing comparatively little use.

The Index was explicitly framed as a monitoring tool rather than a one-off study — a way to track, empirically and over time, claims about AI’s labour-market effects that had mostly been argued from theory or from vendor anecdote. Anthropic and outside researchers published further instalments through 2025 and into 2026, extending the method to survey data, usage-cadence tracking and cross-industry comparisons, making the Index one of the more sustained attempts by a frontier lab to make its own usage data available for independent scrutiny.