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
为什么我把整个产品都建在 Cloudflare 上
大多数 AI 产品跑在 AWS 或 Vercel 上 + 某处的数据库。我把每一层都跑 在 Cloudflare 上——Workers、D1、R2、KV、Vectorize、AI Gateway、Pages。 六个理由它加起来意味着更少的代码和更低的账单，以及两种我不会用它的 情况。
- Eureka
Anti-Sycophancy: A Five-Line Rule That Changed My Roundtables
LLMs are trained to be helpful, which is to say agreeable. In a multi-agent roundtable that is the failure mode. Five lines, added to every persona, brought back disagreement and made the discussions actually useful.
- AI Products
不取悦的教练
大多数 AI 产品在为"用户当下感觉好"做优化。一个教练必须为"用户变得更 好"做优化。这两个目标真的会冲突。从选第二个出发的三个产品决策。
- Agents & Systems
Episodic Memory, and Why Most RAG Is Forgetful
RAG remembers what was said. Episodic memory remembers what happened. The difference is whether your system can learn — or just recite. A recipe for adding episodic memory to a RAG-only system in a week.
- Agents & Systems
大多数人忽略的 Agent 系统的五层结构
人人都把"LLM 加几个工具"叫做 agent。这个词背后藏了太多东西。我把它拆成 五层，每层缺位时的症状写在旁边，结尾给一个 30 秒电梯版。
- Agents & Systems
记忆是底座，不是功能
大多数 AI 产品把记忆当成"撒上去的功能"。好的产品把它当成底座。一个 真正的记忆层需要的六件事，"向量库 + RAG" 给不了你。