Anthropic Economic Index

The Anthropic Economic Index is Anthropic’s recurring study of how its AI model Claude is used across tasks and occupations, launched on February 10, 2025. It sorts anonymized conversations into the tasks listed in O*NET, the US Department of Labor’s occupational database, using Clio, Anthropic’s privacy-preserving analysis tool, and then groups those tasks into occupations.¹ Because the data come from Anthropic’s own product, the Index measures observed use of one AI system, not adoption across employers; its first paper says its data “only paint a picture of AI usage on a single platform”.²

That first paper analyzed over four million Claude.ai conversations and found that about 36% of occupations used AI for at least a quarter of their associated tasks, and that 57% of usage suggested augmentation and 43% automation.² Anthropic reported that 37.2% of conversations fell in the computer and mathematical category, covering tasks such as software modification and code debugging.¹ It has since published six Index reports, the latest on June 26, 2026, with the data on Hugging Face under a CC-BY license; the September 2025 release added geographic and first-party API data.³ The June 2026 report, “Cadences”, switched to daily sampling, broke out chat and Cowork conversations from API traffic, and added first results from the Anthropic Economic Index Survey launched in April 2026.⁴

Two related Anthropic studies build on the data. In March 2026 it introduced “observed exposure”, which combines theoretical language-model capability with usage and which Anthropic calls “a new measure of AI displacement risk”; that is Anthropic’s label for its own measure. In the same paper it reported “no systematic increase in unemployment for highly exposed workers since late 2022”.⁵ In September 2026 its robot study reported that “robots can already perform 74% of physical tasks in the US” but are cost-competitive for 0.3% of job tasks.⁶ These are usage and exposure measures, not counts of jobs lost. MetaRoles sets the Index beside other indices in how we measure AI in roles, and cites its software share in our feature on coding agents and developer productivity.