Claude Leads 26% of Anthropic's Own AI Research

Anthropic has published data. Claude leads 26% of the company's AI research work. That means the model completes most of a task on its own from a high-level request, while a human observes. On more than 90% of research work, Claude does "large chunks of work" under direct guidance. CEO Dario Amodei has proposed making such measurements an industry standard.

The 26% figure is a careful one. These are supervised tasks, not autonomous work. The more interesting number is 90%. Claude already does "large chunks" of research work, which means it is part of the core workflow, not a supporting tool. If other labs follow Amodei's proposal, people building with agents will have a proper point of comparison.