How Anthropic ships Claude with Claude: Separating the Brain from the Hands
Anthropic uses Claude internally to ship Claude, separating agent brain from hands
“having good evals is golden.”
Anthropic engineering and product leads detail how the company dogfoods Claude in its own development workflows, including the Opus 5.5 launch which achieved cost reductions over its predecessor while holding the #1 LLM position for a month. The episode surfaces a key architectural mental model—separating an agent's 'brain' from its 'hands'—and reveals launch complexity like protecting reasoning traces from distillation extraction. This matters because it gives practitioners a rare inside view of how a frontier lab operationalizes agentic development at scale.