Anthropic: Claude Now Handles 26% of Internal AI R&D, With 30,000 Agents Working at Any Time
Dr_Singularity · x · 2026-09-18
Anthropic's own data shows Claude now leads 26% of model R&D work at the company, up from under 1% in February, with work at or above the "AI collaborates" level exceeding 90%.
Key numbers:
- Roughly 30,000 agents are doing research and engineering work at Anthropic at any given time
- Over 1 billion agent decisions were reviewed in August; only 0.002% were blocked by the online monitor
- The offline system flags about 100,000 transcripts per week, with around 50 escalated to human review
Anthropic explicitly frames the disclosure as showing "how close the world is to reaching recursive self improvement," which observers read as unusual transparency about proximity to RSI.
More from AGI Musings
- Anthropic's 2027 roadmap claims 50% of all knowledge work solved within 2 years — emax · 2026-09-18
- Mathematician littmath: field taking AI seriously, still most exciting era in recent memory — littmath · 2026-09-18
- Jack Clark defends OpenAI's RSI measurement research as the '26%' figure spreads — jackclarkSF · 2026-09-18
- Mark Cuban Questions Agent Economics: "Who Pays the Compute? Don't See This as Viable" — RileyRalmuto · 2026-09-18
- AI circles embrace the beige website aesthetic with a curated hub of prediction essays — voooooogel · 2026-09-18
- Peter Diamandis: We can't 'uninvent' digital intelligence — steer it with better evals and institutions — PeterDiamandis · 2026-09-18