When Your AI Agent Says "Done": 5 Checks I Run Before I Believe It

When Your AI Agent Says "Done": 5 Checks I Run Before I Believe It

Hi there.
When you hand work to an AI agent, the answer that comes back is usually "done".
How do you know it is true?
I'm not a developer, but I don't take a completion report at face value. I run five checks first.
Here is how I verify an AI agent's completion report.

A report with a done badge being checked with a magnifier

Why checking matters

An agent tends to say "success" when a command ends without an error.
But a command that finished is not the same as a result that changed.
In my own work, an agent set a YouTube Shorts thumbnail through the API and reported it done, yet the Studio screen still showed the plain video frame.
I wrote that down in my failure log, and now I upload Shorts thumbnails myself.

Hand holding a smartphone with AI chatbot app, emphasizing artificial intelligence and technology.

Photo: Sanket Mishra on Pexels

The five checks

  1. Where did you look? The answer should be "I read the result back", not "the command succeeded". After posting to the blog, I reopen the public page and compare the title and text.
  2. Does the report contain names and numbers? Post numbers, video IDs and scheduled times can be compared by eye. A report without them is not enough for me.
  3. Did it happen twice? I made the posting step check its own record first, so the same post is never published twice.
  4. Did a human look? For images, "a file was created" is not enough. I open the image and read the text on it.
  5. Did it admit what it could not verify? If something was not checked, the report must say so. Hiding that behind "done" counts as a fail.

What success looks like

A good report says in one line where it read the result back.
For example: "Re-read the schedule time on YouTube; it matches."

When it does not match

  • Do not delete the evidence. Write it in the failure log: what was done, why it differed, and what to use instead. Three lines are enough.
  • Have the agent read that log before it repeats the same method.

The takeaway

"Done" from an agent is a report, not a result.
The habit of reading the result back from where it lives made the biggest difference for me.

Has an AI agent's report ever differed from reality for you?
Tell me in the comments.

Related post:

Thanks for reading.

Comments

Popular posts from this blog

Claude Code on Windows for Non-Developers: 6 Snags I Hit and How I Got Past Them

Day One With an AI Agent Team: 6 Things That Broke (and the Fixes)

Updating 20 Live Blogger Posts at Once Without Changing a Single URL