Lilian Weng's New Blog: Harness Engineering and Recursive Self-Improvement
机器之心 · wechat · 2026-07-07
OpenAI's Lilian Weng published a new blog post, "Harness Engineering for Self-Improvement," returning to a high-frequency posting mode less than 10 days after her last update. The article systematically outlines the "Harness layer" situated between foundation models and the real world—the system architecture responsible for orchestrating model thinking, invoking tools, managing context, and evaluating results. It covers design patterns like workflow automation, file-system persistent memory, and sub-agents, alongside frontier research directions such as ACE, MetaContext Engineering, and the Darwin Godel Machine. Weng explores whether Recursive Self-Improvement (RSI) will first occur at the model weight level or within the Harness layer, candidly listing current bottlenecks: ambiguous evaluation metrics, context and memory lifecycle management, the systematic neglect of negative results, diversity collapse, and reward hacking.
Related event: Lilian Weng and Sakana AI Explore Harness Engineering for RSI(4 posts)→
More from coding & agent
- Alex Townsend posts 200 open problems in numerical linear algebra for humans and AI agents — IgorCarron · 2026-09-11
- Kimi K2.8 Preview rolls out: near-K3 coding performance, 1M context for all tiers — teortaxesTex · 2026-09-11
- Looking for a classifier of software engineering task shapes to pick models per task — StewartalsopIII · 2026-09-11
- Steal this idea: prompt-to-hardware where agents assemble custom devices — paraschopra · 2026-09-11
- Model Is the Least Interesting Part: A Guide to Six Core AI Architectures from RAG to Multi-Agent — goyalshaliniuk · 2026-09-11
- Non-coder builds layered memory architecture: 20k tokens tracks a year of agent conversations — matteoianni · 2026-09-11