Free open-source course teaches harness engineering by rebuilding Claude Code from scratch
pauliusztin · reddit · 2026-10-24
An AI engineer who reverse-engineered Claude Code, OpenCode and Pi released a free open-source course on harness engineering: 8 articles, 6 videos and one Python codebase for building a coding agent from scratch.
Key evidence: LangChain kept the same model and changed only the harness, lifting its coding agent from 53% to 67% on Terminal-Bench — from outside the top 30 to the top 5.
The 8 core steps covered:
- Design the harness
- Build the bare-bones agent loop
- Sandbox it locally and remotely
- Apply 4 essential context engineering techniques
- Add subagents as a context engineering technique
- Deploy headless to the cloud: CLI, webhooks, cron
- Build benchmarks to measure it
- Replay failed runs to debug agents
The author argues these foundations power any agent domain, and that teams increasingly build custom harnesses instead of relying on Claude Code or Codex (e.g. Ramp). Sponsored by Modal, Opik and Kitaru; code on GitHub.
More from coding & agent
- Codex CLI's hidden remote-control feature praised as better than Claude Code's — antirez · 2026-10-24
- PWAs can't use foldable screens on the Duo yet, but two Chrome flags unlock the APIs — mitsuhiko · 2026-10-24
- Reddit user asks which no-code AI agent builders actually work beyond the demo — Austin_Ox23 · 2026-10-24
- Dev compares Claude vs OpenAI's Astra: Claude fixes its own mistakes, Astra spins for days — StewartalsopIII · 2026-10-24
- Qwen 27B local RAG on a 3090 ties Opus 5.5 on Korean benchmark — Front-University4363 · 2026-10-24
- Building with AI, drowning in research: when does optimising the process become procrastination? — SJW_Shadow_Monarch · 2026-10-24