This talk presents Agentic Harness Engineering, a closed loop for automatically improving coding-agent harnesses. Component, experience, and decision observability make every edit explicit, evidence-based, reversible, and testable, enabling reliable evolution beyond manual trial and error.