Claude Now Leads 26% of Anthropic's AI Research and Development, Up From Under 1% in February

My take: Anthropic revealed this week that Claude now leads 26% of its internal research and development work, the work that builds the next Claude, up from under 1% in February 2026. It is one of the first times an AI company has published concrete metrics on how much AI does in the internal work that improves the model itself. The company runs over 30,000 agents in parallel, and more than 90% of that work involves the model as an active collaborator.

What matters is not the number itself, but the speed of the change. In seven months, AI went from being an occasional assistant to leading a quarter of R&D work at one of the world's leading AI companies. But no category sits at full autonomy (AL5 on Epoch AI's scale): human oversight remains part of the process.

A necessary clarification: Anthropic measures its own models with its own internal scale. As judge and party, these figures should be read with your own critical judgment, and third-party validation is worth waiting for before treating them as definitive.

If you lead a team or manage projects, the concrete question is this: do you know how much analysis, review, or content generation AI is doing in your operation today, compared to six months ago?