How Managers Can Actually Use AI (Not Just Talk About It)

The Gap Between Talking and Doing

There's a version of "embracing AI" that looks like this: attending a company lunch-and-learn, nodding along, going back to your desk, and opening the same spreadsheet you had open before.

I've seen it a hundred times. The manager who can tell you exactly what a large language model is, who has a strong opinion about which company will win the AI race, who has maybe even tried ChatGPT once to write a birthday card. And who has absolutely not changed a single thing about how they do their job.

That's not using AI. That's being an informed spectator.

The managers who will win with AI are not the ones who attend the most webinars about it. They're the ones who sit down on a Tuesday afternoon with a real problem they need to solve and figure out what the tool can actually do. I want to be specific about what that looks like, because the abstract version of this conversation isn't helping anyone.

Start With a Real Problem, Not a Test Drive

The mistake most managers make when they first try to use AI seriously is they go looking for something to do with it. They'll open Claude or ChatGPT and think "okay, impress me." That's not a workflow, that's a demo mindset, and demos don't change habits.

The approach that actually works is the opposite. Start with the annoying thing on your plate right now, and ask whether AI can help you do it faster or better.

A few weeks ago I had to prep for a performance conversation that I'd been dreading. The person involved had real strengths, some genuine blind spots, and a tendency to get defensive when given direct feedback. I'd written out a rough set of talking points and I wasn't happy with them. They were either too soft or too clinical.

So I gave Claude the context. Not the person's name, not anything sensitive, just the shape of the situation: here's what I need to communicate, here's what tends to go wrong in these conversations, here's the outcome I'm hoping for. Help me think through the framing.

What came back wasn't a script I could read verbatim. It was a different way of sequencing the conversation that I hadn't considered, a reframe on the opening that felt more honest, and two or three specific phrases that were sharper than what I had. I took about 40% of it and threw the rest away. The conversation went better than I expected.

Point being: I didn't use AI to avoid the work. I used it to think harder, faster.

The Actual Use Cases Worth Your Time

Let me be concrete, because vague advice about "leveraging AI" isn't actionable.

Preparing for difficult conversations. This is probably the highest-value use case I've found for managers specifically. You give the model the situation, the person's likely reaction, the outcome you need, and ask it to pressure-test your approach. It will find the holes in your argument faster than most peers will, because it has no social stake in telling you the comfortable version.

Processing meeting notes into action items. If you're recording your meetings, you can run a transcript through a model and ask it to pull out the commitments, the open questions, and the decisions made. This sounds trivial until you've done it after a two-hour strategic planning session where six different threads got pulled and nothing got cleanly resolved. Getting that summary in five minutes instead of spending twenty minutes reconstructing it from memory is genuinely useful.

Writing first drafts of things you hate writing. Job postings, performance improvement plans, team updates, policy documents. These are not documents that require your creative genius. They require a competent, clear first draft that you can then edit to make it real. Let the model produce the competent first draft. Your job is to make it yours.

Thinking out loud with a patient interlocutor. This one took me longer to appreciate. Sometimes you have a gnarly organizational problem and you need to work through it with someone. The problem is that the people in your organization are stakeholders in the outcome, so they can't be fully neutral. A model has no stake. You can think out loud, explore options you'd never say out loud to a direct report, and get pushback that isn't filtered through anyone's self-interest.

What AI Can't Do For You

This matters as much as what it can do.

AI cannot build the relationship with your team that makes feedback land. It cannot read the room in a one-on-one. It cannot sense that someone's energy is off today and adjust accordingly. It cannot make the judgment call about which of your two highest performers to put on the critical project when you can only choose one.

The mistake I see some managers making right now is a new version of an old failure: delegating the thinking because the tool makes it easy to do so. I've seen managers paste a whole performance review into a model and ask it to "write the review for me." Then they sign their name to it without reading it carefully. That's not using AI. That's abdicating your job while getting credit for doing it, and your people deserve better than that.

The tool should make your thinking sharper. It should not replace it.

Building It Into How You Actually Work

The managers I know who are genuinely getting value from AI right now have one thing in common: they've built small habits around it, not big transformations.

They open the model the same way they'd open a notepad. They paste in a draft email they're not happy with and ask what's unclear. They describe a decision they're wrestling with and ask for the strongest argument on each side. They finish a hard week and ask the model to help them think about what they'd do differently.

None of this requires a company AI initiative. None of it requires approval from your IT department. It requires you to spend twenty minutes today actually using the tool on a real thing, instead of thinking about how you'd use the tool someday.

That's the gap between the managers who will look back on this period as a competitive advantage and the ones who will look back and wonder what happened. Not strategy, not vision, not a bold AI transformation roadmap. Just the willingness to pick up the tool and use it on Tuesday.

The Practical Starting Point

If you've read this far and you're still not sure where to begin, here's what I'd suggest.

Pick one thing on your plate this week that you find tedious, draining, or genuinely hard to think through alone. Write two or three sentences describing the situation and what you need. Paste it into Claude or ChatGPT and see what comes back. Don't accept the first output as finished. Push on it, ask follow-up questions, tell it what's missing.

Do that once. See what happens. If it saves you an hour, or helps you crack something you were stuck on, do it again the following week. That's how the habit builds.

You don't need a framework for this. You need to start.