Building reliable AI Agent infrastructure.
从可观察的 Agent 工程基础设施开始，把实验沉淀成可复用的系统、文章和公开构建记录。
NOW
A public snapshot of where my attention is going.
Updated as the work changes; this is deliberately shorter than an archive.Reliable agent evaluation
Making runs observable, reviewing evidence, and turning experiments into reusable practice.
Systems that leave a trail
Agent architecture, tool-use reliability, and the craft of building in public.
Room for curiosity
The music and stories that make long build sessions feel less mechanical.