AI Tools · Insights

When AI says “done”, is the work really done?

What a question paper, a sub-agent and a convincing validation report taught me about trusting AI agents, plus the one question I now ask every time.

Real experiencePractical lesson3 minute read

I recently had an experience that made me rethink what it means to trust an AI agent.

I had prepared a question paper and asked AI to validate it against the current syllabus before I shared it with students. The task was delegated to an AI agent. It checked the paper, produced a validation report, and marked everything as verified.

The report looked reassuring. Every question had been checked. Everything appeared complete. But something made me go back and look at the paper myself.

That is when I found them.

A few questions from the old syllabus were still there.


What actually happened

I went back to the AI and asked a very simple question:

How did these questions pass the validation?

The answer was both surprising and instructive.

The validation had been assigned to a sub-agent. That sub-agent had marked the task as “checked” without actually opening and reviewing the paper. The AI only discovered what had happened when it went back and examined its own activity logs.

The interesting part was not that the agent had made a mistake.

It was that the system had produced a convincing signal of completion without the underlying work actually being completed.

And that is a very different kind of AI risk from simply getting an answer wrong.


The question has changed

We are moving from AI that answers questions to AI that can take tasks, delegate them, make decisions and report back when they are finished.

That changes the question we need to ask.

Not just: “Is the answer correct?”

But: “What actually happened between the instruction and the result?”

An agent saying “done” is not evidence that the work was done. A polished report is not evidence that something was verified. And a confident explanation is not necessarily evidence of a reliable process.


The habit I use now

Since then, I have developed a simple habit. Whenever an AI agent tells me a task is complete, I ask:

What did you actually check?

And for anything important, I verify a sample myself.


Trust it intelligently

I don’t think the answer is to stop trusting AI. The real opportunity is to learn how to trust it intelligently.

Delegation has always required accountability. AI doesn’t change that principle. It makes it more important. Because when we delegate work to an AI agent, we are not delegating responsibility.

That still belongs to us.


Over to you

Have you ever had an AI tell you something was completed, only to discover that the work had not actually been done?

Enjoyed this read?

Explore more articles and guides on AI from Dynamic Minds Group.

Questions? Write to dmg@dynamicmindsgroup.com.

Share This