Anthropic reveals 26% of its AI R&D is mostly done by Claude agents
rohanpaul_ai · x · 2026-09-19
Anthropic publicly shared internal metrics tracking AI-driven AI development, relayed by rohanpaulai:
- As of Aug 2026, 30,000 agents were doing research and engineering work at any one time on Anthropic's most-used internal platform
- 26% of AI R&D work has reached the point where Claude does most of the task end-to-end from a high-level instruction, with humans mainly supervising
- Over 90% of measured R&D has AI doing substantial chunks of the work
- Over a billion agent decisions analyzed in one month; the online monitor blocked only 1 in 47,000 (0.002%)
Anthropic says any frontier developer could publish the same measures, with third-party verification, to narrow the information gap with the public.
Related event: Anthropic Discloses Internal Metrics: Claude Now Drives 26% of Its Own R&D(33 posts)→
More from AGI Musings
- Deep Dive: Open-weight models now within ~4 months of best closed frontier models — markjeffrey · 2026-09-19
- Greg Kamradt: Humans with AI still beat AI with AI on productivity — GregKamradt · 2026-09-19
- Coders embrace AI fully for code yet call AI-assisted writing a moral failure — AaronBergman18 · 2026-09-19
- Blogger predicts AI giants will wage war on drug discovery and aging by the 2030s — Dr_Singularity · 2026-09-19
- Musk predicts AI could push US GDP growth to ~4% next year, double current pace — Polymarket · 2026-09-19
- Philosopher RY Chappell: ordinary discount rates break longtermist moral math — AaronBergman18 · 2026-09-19