Martin Fowler Formalizes Harness Engineering: Agent = Model + Harness as the Core Mental Model for AI Coding
Martin Fowler·high signal
Martin Fowler published a full article on harness engineering — everything in an AI coding agent except the model itself. The key insight: instead of manually fixing AI output (on-the-loop), engineers should improve the harness that produces the output, creating a continuous improvement loop. Harness encompasses context engineering, architectural constraints, and garbage collection. This frames the emerging discipline that makes AI agents production-ready.