On Humans and AI
Blog, translations, things I've built, and thoughts along the way.
GitHub ↗
1,299 contributions in the last two months
Recent Activity
Harness Engineering为你的 agent 构建可协作、可反馈、可约束、可持续迭代的运行环境。Harness design for long-running application developmentHarness engineering: leveraging Codex in an agent-first worldOver the past five months, OpenAI's Harness team built and shipped an internal beta with zero manually-written code using Codex agents—roughly 1M lines across ~1,500 PRs at ~1/10th the development time. This post shares what they learned about designing environments, specifying intent, and building feedback loops for agents.