www.lesswrong.com/posts/LJbKwctaioqp2Hi4b/superintelligence-this-christmas
1 correction found
Claude has completely taken over 26% of all R&D.
Anthropic did not report that Claude had fully taken over 26% of all R&D. Its published metric says Claude "leads" 26% of Anthropic’s AI R&D work, with human supervision, and explicitly says Claude is not fully autonomous in any measured subset.
Full reasoning
Anthropic’s measurement page draws two distinctions that the post collapses:
- The scope is AI R&D, not all R&D. Anthropic says it built an index of how much of Anthropic’s AI research and development is performed by Claude.
- "Leads" does not mean fully autonomous or "completely taken over." Anthropic defines its AL4 category as AI completing most of a task from a high-level prompt while a human supervises, and it separately states that Claude is not operating fully autonomously for any measured subset of AI R&D work.
So the official claim is narrower and weaker than the post says: Claude "leads" 26% of AI R&D work, not 26% of all R&D, and not with complete takeover.
Anthropic’s page also says the ~30,000-agent figure applies to agents working "at any one time in our most-used internal platform," not necessarily to all company systems taken together.
1 source
- Measurements for understanding the pace of AI development inside frontier labs | Anthropic
What we measured. We built a prototype index of how much of Anthropic's AI research and development (R&D) is performed by Claude... As of August 2026, Claude is not operating fully autonomously for any measured subset of AI R&D work. Claude "leads" 26% of Anthropic's AI R&D work.