Learning学习

Lately I have been working out how to lead AI agents through to a finished result.

That has taken me into application development, automated workflows, research and deployment. I usually start with a concrete project, then learn what I need as I encounter each problem.

Generating a result is only the first step. Someone still has to judge whether the requirements are clear, the sources are reliable and the result solves the problem. When a model says it is done, I check what it has actually completed.

I define the requirements, choose an approach and review the result, while agents carry out the work and implement the code.

AI is a powerful lever: it lets me take on work well beyond my original field. I am willing to try ideas in public and improve them using the results and feedback. You can find several of those projects in Things Made.

这段时间我一直在研究，怎样带着 AI Agent 把事情做完。

这让我接触到应用开发、自动化工作流、资料研究和部署。我通常从一个具体项目出发，遇到问题再学需要的东西。

生成结果只是第一步。需求有没有说清楚，资料和来源靠不靠谱，做出来的东西能不能解决问题，都需要人来判断。模型说完成了，我还要亲自看看它做到了哪一步。

我负责明确需求、选择方案和检查结果，组织 Agent 完成执行与代码实现。

AI 是很大的杠杆，让我能做原来专业之外的很多事。我愿意公开尝试，也会根据实际结果和反馈继续改进。作品集里有几个这样做出来的项目。