Claude now leads 26% of Anthropic's AI R&D while 30,000 agents run concurrently
krishnan · x · 2026-09-19
Anthropic says Claude now "leads" 26% of its AI R&D work and collaborates on over 90% of it. But krishnan argues it's not a true autonomy milestone: Anthropic admits no measured category is fully autonomous.
The more important shift is operational. In August, Anthropic ran roughly 30,000 agents doing research and engineering work concurrently. Every proposed action was screened — across more than a billion decisions, only about 1 in 47,000 was blocked.
The core insight: once cognition gets cheap, attention becomes the scarce variable. The valuable people are no longer just researchers producing answers, but harness builders who define targets, preserve context, and catch subtle errors — the unit of management is now a fleet of workers generating experiments, patches, tests, and decisions faster than humans can inspect them.
Related event: Anthropic discloses Claude now drives 26% of its own R&D(39 posts)→
More from AGI Musings
- Anthropic publishes three public metrics to track how fast AI is building AI — minchoi · 2026-09-19
- MIT-shared paper argues LLMs can't be conscious without brain-like analog computation — burny_tech · 2026-09-19
- AI hallucination about Chinese nuclear parts nearly triggered US military action, report says — polymute · 2026-09-19
- Six years from GPT-3 to safety incidents — robotics may need only months — jiaxinwen22 · 2026-09-19
- Beyond human mathematics: what kind of AI can even ask new questions? — teortaxesTex · 2026-09-19
- Cybersecurity freshman asks: why is recursive self-improvement actually scary? — n0thingT0S3eHere · 2026-09-19