Ten harness patterns that you can apply to a real repo

Harness Engineering is an approach that improves AI coding agents by building better systems around them—including documentation, context loading, validation, and feedback loops—rather than relying solely on better prompts. The authors share ten practical patterns tested with tools like Codex and Claude, organized into three layers: making the repository an agent-readable knowledge base, loading context just-in-time for tasks, and using validation and feedback to prevent recurring mistakes. These patterns help engineering teams achieve more predictable delivery with less time spent on context reconstruction and agent error correction.