The complete harness engineering roadmap: how OpenAI shipped ~1M lines with zero hand-written code
blaizedsouza · x · 2026-11-11
A comprehensive learning roadmap for Harness Engineering — the discipline behind OpenAI shipping 1M lines with zero hand-written code and Anthropic running agents for hours.
Start here (6 reads): OpenAI's Harness Engineering post, Mitchell Hashimoto's AI adoption journey, LangChain's Anatomy of an Agent Harness, Martin Fowler's practitioner view, and two Anthropic articles on long-running agents. Ready harness: pip install deepagents.
Real cases: 3 OpenAI engineers, 1M lines; Anthropic's initializer→progress-file→feature-list flow; planner→generator→evaluator architecture; Fable 5.2 multi-agent teams coordinating 1,000 agents through one shared database; RHO lifting SWE-Bench Pro from 59% to 78% in one pass.
Papers: Meta-Harness (same model shows up to 6x gap from harness alone), Code as Agent Harness, RHO, and a source-level dive into Claude Code.
Optimal order: read the OpenAI post → the Anatomy → build one loop with Deep Agents → add a progress file and feature list → add an evaluator agent → measure on Terminal-Bench → then papers and auto-optimization. Includes a 14-lecture free course and curated GitHub repos.
More from coding & agent
- Stop burning tokens: a Claude Code orchestrator prompt that routes tasks to cheaper models — TawohAwa · 2026-11-11
- Amazon paper: multi-agent setups burn 1.4-4.7x tokens for small gains; single agent is the default — rohanpaul_ai · 2026-11-11
- Dev shows off building a retro FPS entirely via Codex Mobile — 503 tests passing — Dimillian · 2026-11-11
- Explosion launches Ellf directory: 52 NLP templates plus multi-agent annotation workflow — _inesmontani · 2026-11-11
- Meme: 'Stochastic AI Agent Architecture' Is Just a Markov Chain in Disguise — PrasoonPratham · 2026-11-11
- OpenCode adds Gemini 4 Carbon and Claude Fable 5.2 support — dean_rie · 2026-11-11