Anthropic Economic Index reports on how Claude use changes as users gain experience

Anthropic published a report, the latest instalment of its Economic Index series, examining how Claude use changes as users gain experience with it. The analysis drew on roughly a million Claude.ai and first-party API conversations sampled over one week in early February 2026, processed through Clio, Anthropic’s privacy-preserving conversation-analysis system, rather than on user interviews.

Users who had used Claude for six months or longer showed about 10% higher success rates in their conversations than newer users, a gap that held after controlling for task type. Long-tenured users had also broadened what they used Claude for: the share of traffic going to the ten most common task types fell as usage matured, and users increasingly turned to it for higher-education-level work — described in the report as tasks demanding roughly a year more education for each year of platform tenure — moving away from casual uses toward things like research and startup fundraising work. Experienced users were also more likely to select Anthropic’s most capable model, Opus, for demanding tasks.

On autonomy, the report found a more modest shift than the “users increasingly delegate to AI” narrative might predict: high-tenure users were not markedly more likely to hand off full responsibility through directive, delegate-and-walk-away patterns, and instead favoured iterative collaboration — checking and refining outputs rather than issuing a task and stepping back. Automation-style usage patterns fell in the API data even as augmentation-style usage rose slightly on Claude.ai.

The report is one of a running series Anthropic has used to track aggregate patterns in how its models are actually used, feeding into the broader debate over how quickly AI assistance is being adopted for skilled economic work versus casual tasks.