One of my sons and I were talking the other day about whether AI makes us less capable of thinking for ourselves.

My answer was: it depends on how you use it.

Before I go on, I should explain the AI part of my tech stack. In my current setup, I work with an AI assistant I call Guy (pronounced GEE) through Codex. Much of that interaction is spoken conversation at different times throughout the day. I talk through an idea, ask questions, challenge a response and refine what I’m trying to accomplish.

That’s important context for my answer.

If you ask AI for an answer, copy it without reading it and move on, you haven’t necessarily learned anything. You may not even know whether the answer is right.

My experience has been different. I’m having more conversations about my work instead of keeping everything in my own head. I feel more confident, and I think my communication has improved.

I work independently. In the past, I could spend hours in my office solving problems on my own. Now, I have what I’ve started calling a virtual water-cooler companion. Talking things through often helps me discover what I’m really trying to ask.

Sometimes I learn something new. Sometimes I just get something done faster.

Today, while updating a client’s business-intelligence workbook (Power BI), I noticed that refreshing roughly 1.2 million records from SQL Server took over a minute. Nobody had complained, but I wondered how that wait felt to people refreshing it every day. I wanted to improve their experience rather than simply accept it.

Did they need all that history for their regular analysis? Not likely. We had applied a filter two or three years earlier, and nobody had asked for the older history it excluded. But choosing the right cutoff was still a question for the stakeholders who use the information.

I couldn’t remember how far back the data now went. Before asking anyone to choose a shorter period, I wanted a simple summary by year to help frame the question.

Guy quickly helped me write a SQL query to produce that summary. Could I have written it myself? Yes. I’ve been doing this kind of work for decades. But I wanted the evidence so I could move on to the decision.

I checked the query, ran it and reviewed the results: the history went back to 2020. Guy then helped draft a clear email asking stakeholders how far back they needed to go for their regular analysis. Would loading only 2023 onward meet their daily needs, or would another starting year make more sense? The older data would remain available, and I could quickly relax the filter when needed.

In a short time, we’d moved from a concern to evidence to a proposal people could evaluate. I’m now waiting for their feedback. We haven’t changed the filter yet; that decision needs their input.

I don’t need every interaction to become a lesson. I do need to stay involved enough to know what I’m accepting.

This example comes from my own workflow. Someone else’s useful conversation might look quite different. I’m interested in showing people what’s possible and letting them consider where it might fit their work. I don’t need to convince everyone.

But for me, having somewhere to think out loud has made a difference.

Are you talking with AI, or mainly typing questions into it? If you’re having spoken, back-and-forth conversations, what are they helping you work through? Or are you reading this thinking, “Talking to AI? I’m not ready for that conversation yet.”

I’d like to hear that perspective too.