Writing

让判断过程
可以被讨论。
Make the reasoning
discussable.

文章记录对 Agent、工程方法和人机协作的独立思考。每篇明确内容状态与公开性边界,并以中文完整呈现。The articles record independent thinking on agents, engineering methods, and human collaboration. Each one states its content status and public boundary, and is presented in full in Chinese.

Independent research

从两个问题开始。Start from two questions.

先讨论边界如何移动,再讨论开发 Agent 如何形成可验收的循环。First, how the boundary moves; then how agent development forms an acceptable loop.

独立研究 · 已发布Independent research · Published

Agent systems · boundary design

Agentic 的
边界悖论
The agentic
boundary paradox

为什么企业 Agent 项目的关键矛盾不在模型能力,而在自主边界只能随实施逐步显现。Why the key contradiction in enterprise agent projects is not model capability, but an autonomy boundary that only becomes visible during implementation.

Agenticworkflowhuman-in-the-loop
阅读文章Read the article
独立研究 · 已发布Independent research · Published

Agent development · evaluation loop

Loop
Engineering

不是把提示词写得更长,而是将 Agent 开发拆成可观察、可评估、可迭代的工程循环。Not writing longer prompts, but breaking agent development into an engineering loop that can be observed, evaluated, and iterated.

Skill developmentevaluationfeedback loop
阅读文章Read the article