AI now leads 26% of Anthropic's AI-model R&D work, company data shows

AI companies sell tools that take on other people's work. Anthropic has now published a count of how much of its own work those tools lead.

In a post on measuring AI development inside frontier labs, Anthropic rates its model research and development tasks on an automation scale from Epoch AI. At the level it calls "leads," AI "can complete most of the task end-to-end from a high-level prompt, while the human supervises." The company's chart, also published by The Wall Street Journal on October 30, puts that share at 1% in early 2026, then 3%, 12%, 14% and 22%, and 26% in August.

None of the work runs on its own. The company says "Claude is not operating fully autonomously for any measured subset of AI R&D work." The Journal's chart note puts it plainly: "As of August, 0% of AI R+D is fully autonomous."

Anthropic's index weights each kind of task by the staff time spent on it. It does not report hours saved, output or cost. It measures who leads the work, not what the work is worth.

Anthropic says it plans to "embed independent third-party evaluators" at the company. Until those checks report, the 26% is the company's own count of its own work.