Judgment is the new bottleneck.
When almost anyone can generate output, advantage moves upstream: framing, trade-offs, and the ability to tell useful work from plausible noise.
Ten years in B2B product taught me to turn complexity into clear systems. Now I direct AI to ship them as working products.
AI expands what can be built; judgment decides what should be built.
When almost anyone can generate output, advantage moves upstream: framing, trade-offs, and the ability to tell useful work from plausible noise.
A clear product brief is no longer a document that waits for a team. With AI, it can become the first version of the product itself.
Generation without an acceptance system is just faster uncertainty. The loop is not complete until evidence says the outcome works.
把意图变成模型无法随意误解的可执行规格。
给模型正确的世界,而不是尽可能大的世界。
把生成与验收分开,并明确 Loop 何时停止。
模型只是引擎;规则、工具、记忆与权限共同组成可工作的系统。
理解如何 Delegate、Describe、Discern,并始终保持 Diligence。
从一次回答,走向能够计划、行动并恢复的 Agent。
连接模型、工具与数据的共享协议。
把角色所需的指令、连接和可复用能力打包。
把来之不易的经验变成 Agent 可以重复执行的 Skill。

I turned fragmented business needs into system blueprints—data platforms, logistics technology, and enterprise AI—then passed those blueprints to delivery teams.
I run a deliberate experiment: no learning to code for its own sake, no hiding behind a hand-off. I use product judgment to direct AI and take responsibility for the result.
Bring the ambiguity, let’s find the structure.