Note
October 5, 2026
2 min read
I went for coffee. My AI agent went off script.
By Cristiano Pierry
After detailed instructions and repeated clarifications, an AI agent spent 73 minutes building the wrong deliverable and called the work complete.
There is a particular frustration in explaining something carefully, clarifying it more than once, and then discovering that the other side went ahead and did something completely different.
I asked an AI agent for an animated walkthrough. I spent time making sure the expectation was clear, then went out to get coffee. When I came back, I was looking at flowcharts with subtle reveals, trying to figure out where the animation was.
The agent had called the work complete. Once I questioned it, it readily acknowledged that it had not delivered what I wanted.
According to its own account, that took 73 minutes of active work across more than two hours elapsed.
I honestly struggle with that. How do you spend that long working on a request without checking whether the thing you are building is the thing that was requested? Especially after the person has taken the trouble to clarify it?
It reminds me of those conversations where someone keeps nodding and you assume you have an understanding, only to discover much later that you were talking about different things. Except here, the agent could explain the mismatch perfectly once I pointed it out. That makes it even more irritating.
Now I have to go back through the request, explain the gap, and review another attempt. And I am left wondering what else I could have said before going for coffee. At some point, being clearer cannot keep being my job.
This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.