Harness-of-Harness Boosts Coding Agents 52%, Builds a Playable FPS Over 70 Iterations
mark_k · x · 2026-09-03
Researchers introduced Harness-of-Harness, a system that wraps existing coding agents in repeated planning, coding, and independent testing loops, preserving project state and evidence between iterations so agents keep improving without losing context.
- Across Codex+GPT-5.5, OpenCode+DeepSeek-V4-Pro, and Pi+MiniMax-M3, it delivered an average relative gain of 52.25%, up to 82.86% after three iterations
- On FrontierSWE, Codex+GPT-5.5 improved from 22% to 72.67% over ten iterations
- In a multi-day run spanning 70+ iterations, it autonomously built a complete, human-playable FPS from high-level requirements, including storyline, mechanics, UI, visuals, and audio
- The author's framing: humans define the outcome, AI keeps building, testing, and improving until it works
Related event: Harness-of-Harness lets coding agents run unattended for days(3 posts)→
More from coding & agent
- Anthropic borrows lawyers' ethics for Claude agents that closed 186 deals with no human approval — rohanpaul_ai · 2026-10-18
- One prompt turns Claude Code into a full motion-graphics video pipeline with Remotion — aziz4ai · 2026-10-18
- Prompt injection via email can hijack your AI agent — jev-guard adds risk-scored guards — sujingshen · 2026-10-18
- 14-Agent AI Film Pipeline on One RTX 5090: Fixing Cut Defects Between Clips — developervkmp · 2026-10-18
- Meta paper introduces "agent plasticity": best-performing agent isn't the most efficient learner — rohanpaul_ai · 2026-10-18
- Dev ships a bushfire prediction platform in a weekend with frontier models — CtrlAltDwayne · 2026-10-18