Boris Cherny: We Cut 80% of Claude Code’s Prompt

Claude Code's prompt reduction illustrates why instructions need retesting when the underlying model changes. Its creator, Boris Cherny, says Anthropic cut 80% of the system prompt because Opus 5 already handled behaviors that earlier instructions had corrected. In the interview transcript, Cherny describes removing the prompt and then restoring it line by line to measure each line's effect, adding an instruction back only after seeing the model repeatedly stumble on the same thing. Some prompts remain useful for the product, while most of the surrounding harness code is about safety and permissions. This supports testing whether old instructions still solve observed problems, rather than treating a shorter prompt as an improvement by itself.