Redditor adapts Google's Harness² into an offline self-improvement loop for Claude/Codex/Gemini agents
cobresalino · reddit · 2026-11-02
Reddit user cobresalino is adapting Google's Harness² approach into an offline improvement loop for a Claude/Codex/Gemini agent setup: learn from session failures, test changes locally, keep what helps, and roll back what breaks.
The post asks whether anyone has pushed this beyond a demo — what actually improved, and how it was measured — inviting repos and failure stories. It highlights a real gap in current agent engineering: making agents genuinely get better with use (autopoiesis).
More from coding & agent
- LLM confidently gives wrong refund policy answer; developer shows 4-step RAG fix — _jaydeepkarale · 2026-11-02
- Dev builds a MetaHuman alternative for three.js in a weekend with Claude — majidmanzarpour · 2026-11-02
- AI agent operates FileMaker from plain English, narrating every click in a cloned voice — petrusenko_max · 2026-11-02
- DeepSeek says it's betting on scaling across agent harnesses, pledging per-generation support — teortaxesTex · 2026-11-02
- What actually happens in the 3 seconds before your chatbot replies: tokens to RAG — Ok_Translator_227 · 2026-11-02
- Video-to-Game Workflow: Video Model Drafts the Gameplay, Claude Builds It in Three.js — OdinLovis · 2026-11-02