Key takeaways
Anthropic discloses Claude’s role in building the next AI: by August AL4 (~AI-led) work was ~26% (from under 1% in February), with 90%+ at least AI-collaborative; ~30k agents on internal platforms made 1B+ decisions in August (~0.002% blocked by oversight). Alongside, Claude Code Projects ships Coordinator + parallel branches + shared memory. Core tension: as AI builds AI faster, can humans still see the pace and keep the kill switch.
