Original Summary

I keep going back and forth on this.<p>Traces and logs tell me what the agent did: every tool call, retry, token, and step.<p>That&#x27;s useful when something breaks. What I&#x27;m less sure about is whether that tells me the agent actually did the right thing.<p>A run can look clean and still miss what I wanted. Or it can &quot;finish&quot; without the outcome really being done. So finished and succeeded aren&#x27;t the same thing.<p>I haven&#x27;t made up my mind how to check for that, but I&#x27;m experimenting a lot with extracting intents from my agent and correlating among them.<p>Curious how you do it:<p>- Do you capture intent and outcome explicitly (e.g. giving a tool to the agent to &quot;declare&quot; it), or infer them after the fact from the traces?<p>- What kind of &quot;questions&quot; do you &quot;ask&quot; your traces?

  • 情报分类：工作与职业机会
  • 分类依据：内容涉及招聘、求职或职业发展
  • 信息来源：Hacker News 新项目
  • 发布时间：2026/10/19 23:11:53