February 2026: Claude led less than 1% of Anthropic's own AI R&D. August 2026: 26%. Claude now does over 90% of the work. And honestly it's a little terrifying that we are only seeing this now, because nobody outside these labs has had any way to check. – 30,000 agents are running research and engineering at Anthropic right now. – Over a billion agent actions reviewed in one month. 1 in 47,000 got blocked. Think about that ratio at scale. –And only 6% of AI R&D compute goes to safety…