mind
A working notebook on AI, agents, and what I keep getting wrong.
Not a tutorial site — a judgment site. Field notes from building AI-native software, mostly out loud, often wrong, occasionally useful.
Recent writing
- Patterns
2026 Agent 框架图景 —— 以及我自己实际站在哪里
五个主流 agent 框架，每个真正解决什么，我自己的多智能体系统在这张地图 里位于哪里，以及"我们走的路对吗？"的诚实答案。
- Eureka
一个 Agent 架构的诚实评估：哪些会留下、哪些不会
大多数架构文章以"作者被自己说服"结尾。这一篇相反：我自己的五层 agent 架构是什么、不是什么、能吸收什么、哪些假设我认为活不过下一次 paradigm shift。
- Patterns
跨产品 Agent 系统的参考架构
五层 stack 不是"我怎么建了一个产品"——是给我未来每个 AI 产品的可复用 底座。共享底座 + 产品特定上层。Amber、Mumu、和元系统本身都坐在同一个 地基上。
- Patterns
Why I Built on Cloudflare Top-to-Bottom
Most AI products run on AWS or Vercel + a database somewhere. I run every layer on Cloudflare — Workers, D1, R2, KV, Vectorize, AI Gateway, Pages. Six reasons it adds up to less code and lower bills, and the two cases where I would not.
- Eureka
反谄媚：改变了我所有圆桌的五行规则
LLM 被训练得"乐于助人"，也就是"好说话"。在多智能体圆桌里，这正是 失败模式。五行规则——加在每个 persona 上——把分歧拿了回来，让讨论 变得真的有用。
- AI Products
The Coach Who Doesn't Flatter
Most AI products optimize for "user feels good." A coach must optimize for "user gets better." The conflict is real. Three product decisions that flow from picking the second.