http://www.heyuan110.com/tags/sub-agent-architecture/Recent content in Sub-Agent Architecture on Bruce on AI EngineeringHugoenMon, 30 Mar 2026 10:00:00 +0800http://www.heyuan110.com/posts/ai/2026-03-30-harness-engineering-guide/Mon, 30 Mar 2026 10:00:00 +0800http://www.heyuan110.com/posts/ai/2026-03-30-harness-engineering-guide/<p><img src="http://www.heyuan110.com/posts/ai/2026-03-30-harness-engineering-guide/cover.webp" alt="Harness engineering architecture — layered systems wrapping an AI agent core with guardrails, feedback loops, and monitoring" loading="lazy" decoding="async" fetchpriority="low" width="1200" height="630" /> </p> <p>This is <strong>Part 1</strong> of the Harness Engineering series. Part 2 goes deep on <a href="http://www.heyuan110.com/posts/ai/2026-03-31-harness-claudemd-guide/">CLAUDE.md best practices</a>. Part 3 covers <a href="http://www.heyuan110.com/posts/ai/2026-04-13-harness-subagent-architecture/">sub-agent architecture and model routing</a>.</p> <p>I have been running an AI coding harness in production for 60 days — the pipeline that writes, illustrates, and distributes posts for this blog. The single most surprising finding from that experiment: <strong>the model is the least important part of the agent.</strong> Not irrelevant, but far from dominant. Upgrading the writer from Sonnet to Opus lifted blind-rated output quality by about 5%. Rewiring the harness — routing, context boundaries, sensors — cut end-to-end cost by roughly 60% while quality went up, not down.</p>