Anthropic disclosed on September 18, 2026 that its Claude model is now helping build its own successor, leading roughly 26% of the company’s model research and development work end-to-end under human supervision, up from zero in February. The company said approximately 30,000 agent instances were doing research and engineering work as of August, and it launched “smart reports,” a new Claude Enterprise beta feature that analyzes team usage, cost and friction points. Anthropic is calling for the AI industry to share standardized metrics on AI-assisted development as models take on more of their own R&D.