a retrospective on making claude code from our early team!

Anthropic's retrospective on Claude Code collects team interviews recorded February to May 2026 and edited. It traces the tool to a demo called Claude CLI, which Boris Cherny, head of Claude Code, says took about two days to build. He says nobody understood it at first, but a colleague soon told him it was doing his coding.

Cat Wu, head of product, says she joined in summer 2024 and used the demo to build reinforcement-learning environments, finding it made her much faster. She says she sent Cherny paragraphs of feedback, he often replied that he had landed a fix, and he invited her to join. The short link is useful as a first-person account of that early feedback loop, but it is the team's own telling, not independent validation.