Lilian Weng argues the agent harness may matter as much as the model itself
sudoraohacker · x · 2026-08-04
Main points
- Lilian Weng’s new long-form post argues that the harness around a model can matter as much as the model itself for self-improvement.
- It breaks harness engineering into patterns such as:
- workflow automation
- file systems as persistent memory
- sub-agents and backend jobs
- The post uses coding agents like Claude Code and Codex as case studies for how modern agent systems orchestrate execution, tool use, context management, artifact storage, and evaluation.
- It frames this as part of the broader RSI/self-improvement loop: improving the deployment system and training pipeline can accelerate the next generation of models.
- The article also outlines open problems such as harness optimization, context engineering, workflow design, evolutionary search, and joint optimization with model weights.
Related event: Lilian Weng: Agent Harnesses Are as Crucial as Models(2 posts)→
More from coding & agent
- Reverse Engineering Reveals Hidden Code in Claude Code Targeting Chinese Users — aftahi_ai · 2026-08-04
- Xberg: Local Document Reading Tool for Agents Supporting 101 Formats — Goldziher · 2026-08-04
- Hermes Agent Optimized: Boosting Small Model Efficiency via 250k Conversation Analysis — max_paperclips · 2026-08-04
- Develop with Two Prompts, Deploy with One: AI Coding Integrated with Stripe — jeff_weinstein · 2026-08-04
- Berlin AI Meetup: Framework-Free Agents and Applied NLP — hugobowne · 2026-08-04
- Qwen 3.8 Early Tests: Good at Agentic Coding, But Struggles with Long Loops — bindureddy · 2026-08-04