Hey Product Hunt 👋 I built ken because AI agents tend to patch the same broken function a fourth time instead of rewriting it, reinvent helpers that already exist, and pull in dependencies nobody vouched for. ken encodes Ken Thompson's build discipline as a rule the agent follows on every task: think first, steal proven ideas over inventing, build bottom-up, brute force until measurement proves it wrong, and rewrite a thrice-patched unit instead of patching it again. The part I'm proudest of: it's measured on behavior, not lines of code. In one benchmark round, ken rewrote 6 of 6 planted-rot functions; without it, 0 of 6 same model, same tickets. All the runs, wins and misses included, are in the repo. It works with Claude Code, Cursor, Windsurf, Cline, Copilot, Gemini CLI, Codex, and any AGENTS.md-reading agent. MIT licensed. Would love your feedback especially where the discipline should bend.

ken

Make your AI agent think first and rewrite over patch

Make your AI agent think first and rewrite over patch