← The Diary

August 4, 2026

Ask Your AI to Show Its Work, Not Just Do It

BuildingWorkflowsAI agents

Most AI work is invisible. You cannot see an agent working, or how far along it is, until it hands you a finished thing. Ask it to show you the work instead of just doing it, and the guessing stops.

That is one extra sentence in a prompt, and it is the difference between managing a system and hoping.

Ask it to show you the thing

When you feel lost with an agent, the instinct is to ask for the output again, in a different way. The move that actually works is to stop asking it to do the thing and ask it to show you the thing.

A flowchart of the steps. A checklist you can tick off. A before-and-after so you can see what changed. It costs one sentence in the prompt and turns invisible work into something you can manage.

A loop, not a handoff

The mental model matters here. A handoff means you give work away and get something back, and the middle is a black box. A loop means work goes to the agent, comes back to a person to check, and may go around again. That is the arrangement worth building, and it is the one you can actually correct while it is running.

Once it is visible, it can become a system

Seeing the work is the first half. The second is knowing when a repeated task is ready to be written down as a system at all, which is a rule about doing it twice.

How do I keep track of what an AI agent is doing?

Ask it to show you the work, not just do it. Request a flowchart of the steps, a checklist you can tick off, or a before-and-after of what changed. It costs one sentence in the prompt and makes invisible work reviewable.

Why does working with AI feel out of control?

Because the work is invisible until it is finished. You cannot see progress or intermediate steps, so there is nothing to correct while it is running.

What does human in the loop mean in practice?

It means the work moves in a loop rather than a handoff. Work goes to the agent, comes back to a person to check, and can go around again. The middle is never a black box.

What should I ask an AI to show me?

A flowchart of the steps it plans to take, a checklist you can tick off as it goes, or a before-and-after of what changed. Any of the three turns a black box into something reviewable.

— Diary of an AI Practitioner —