That Computer Task You Keep Putting Off? Start Here.

You probably have a computer task you’ve been meaning to get to. Old records to review. Information to compare. Something that matters, but never quite makes it ahead of the work right in front of you.

Before you decide it needs another afternoon at your desk, it’s worth asking whether you need to do every part of it yourself.

At my September 17 workshop, I kept coming back to one question:

Is this something that can be done on a computer?

Not because a yes means AI can take over. It means there’s something worth looking at.

That question helped one participant see a possibility in a database they’d been using for years.

The records were already there. The opportunity wasn’t obvious.

The participant had years of business records and suspected there were duplicates. They asked whether AI could help.

We talked through what an agent might do: review records, look at the notes, and flag possible duplicates for someone to check.

That’s a different starting point from asking AI to clean up a database. You’re giving it a piece of the preparation, not permission to decide which business records should disappear.

Imagine getting a review list that puts two possible matches beside each other and explains why they were flagged. The person who knows the business could then check the evidence instead of starting with the entire database.

That’s the opportunity we discussed. We did not complete a database cleanup in the workshop, and I don’t have measured time savings to report from it.

But the conversation had moved somewhere useful. A familiar problem had become a specific piece of work we could consider testing.

Think about what you’d hand to an assistant

Part of what opened up that thinking was setting up Codex projects and talking about an agent more like an assistant, intern, or new employee.

You wouldn’t hand a new employee years of records and say, “Fix it,” then walk away. You’d explain what the records mean, what you’re looking for, and what they need to bring back before changing anything.

The same questions make an AI workflow easier to see.

What information would the assistant need? What would you show them as an example? Which calls would they need to bring to you?

For the database idea, even the word “duplicate” needs some attention. Two similar names might refer to different customers. Two records with different names might belong together. Those are examples of the rules we’d need to clarify, not findings from this participant’s database.

The business knowledge still matters. Giving an agent a task doesn’t remove the need to explain it.

It gives you a reason to get that knowledge out of your head and into instructions someone else can work from.

Start with a review list, not a delete button

For a first test like this, I’d use only data the business permits us to use, keep a backup, and work with a small read-only sample. We’d agree on what counts as a possible duplicate and check the flagged records together. Any merge or deletion would require human approval.

That’s enough to begin learning whether the approach is useful without handing over the whole database.

We also discussed connecting systems, pulling information together, and preparing material for decisions. Those were possibilities to explore, not systems we had deployed.

The participant said they intended to send their more technical right-hand person to a future class. To me, that was an encouraging sign: they could see work worth exploring with someone who understood their systems.

Bring the task, even if you don’t know the tool

You don’t need to arrive knowing how to build an agent. Start with the work you can explain.

Maybe it’s checking records. Maybe it’s comparing two files before making a decision. Maybe it’s gathering information that you then have to sort through yourself.

If it’s done on a computer, we can look at which parts need your judgment and which parts might be prepared for you. Sometimes AI fits. Sometimes a simpler process or a small tool makes more sense.

That’s what my 90-Minute Workflow Optimization Session is for. Bring one recurring task you own. We’ll clarify the work and its boundaries, then work toward a useful first version when the scope allows, or a clear implementation blueprint.

You don’t have to solve the whole business. Bring the task you keep putting off, and let’s see what a sensible first step looks like.