A CEO client asked me recently whether AI was making her dumb. She said it plainly, the way you’d report a symptom. “I’m noticing I feel less sharp these days and I can’t quite put my finger on why.” Dumb is an absurd word for this woman. She runs a company most people spend a decade trying to build. But it was her word, and she’d been carrying it around long enough to say it out loud, so we sat with it. Two things came out.

The first was ordinary. She’d started reading her AI’s first take on a hard problem, nodding, and moving forward, without checking whether she’d have landed there on her own. Small thing, hard to catch in the moment, easy to see once you’re looking for it.

The second mattered more. In the name of not bothering people, she’d started routing her hardest thinking, about people, about culture, about who on her team was struggling and why, through AI instead of through the two or three people who used to help her think about exactly that kind of problem. She thought she was being considerate. What she’d actually done was take the thinking that used to get richer with other people in it and do all of it by herself.

She wasn’t getting dumber. She was getting quieter, and it felt like dulling because it was.

Here’s the decision underneath both of those, and it’s one you’re making dozens of times a week whether you’ve noticed it or not. Which parts of this do I hand to the machine. Which parts, and which people, do I keep in the room.

Most of that decision is easy and correct. The email gets drafted by AI because that’s faster. The summary of a document you don’t need to read line by line gets handled the same way. The first pass at a strategy doc, the version of a message that just needs to exist so you can react to it, none of that is a loss. It’s correctly identifying what didn’t need you, or anyone else, in the first place.

But my client’s story is a second version of that same handoff, and it’s easier to miss. It isn’t about who drafts something. It’s about who you stop consulting. She didn’t stop making decisions about her people. She stopped bringing in the humans who used to help her make them, because the machine was faster, always available, and never seemed put out by being asked twice.

This isn’t a decision anyone makes on purpose. It happens by increments. The draft of a hard message gets written by AI because it’s easier to edit than to create. The performance review gets structured by AI because the blank page is uncomfortable. The read on a tense team dynamic gets run past AI instead of past the colleague who actually knows the people involved, because pinging a colleague feels like an imposition and opening a chat window doesn’t. Each of those is defensible on its own, and taken alone, none of them looks like a line being crossed.

Then one day you notice you haven’t originated a hard call in months. You’ve only ever edited one, or had it confirmed by something that has never met the people involved.

Here’s the test I’d offer, and it has two parts. The first is the one you’d expect: does the fact that I did this myself change what it means to the person receiving it. A quarterly summary, no. A hard conversation about someone’s future at the company, yes. The words might come out the same either way. The weight doesn’t.The second part is the one my client’s story adds. Is there someone I used to want in the room for this, who I’ve quietly stopped asking. Not because they stopped being useful. Because asking a person costs something, a small tax of time or awkwardness or the risk of looking unsure, and asking AI doesn’t. That tax was never a flaw in the system. It was often the point. It’s what kept you talking to the two or three people who turned your thinking into something better than what you walked in with.

There’s something else she lost that neither of us named right away. Thinking with someone isn’t only a better way to think. It’s company. Working a hard problem with a colleague, both of you unsure, both of you invested in getting it right, is one of the few things at work that makes people feel like they’re building something with each other rather than near each other. Take those hours out of a week and the work still gets done. It just gets done alone.

That’s a large part of what felt like dulling to her. She was solving as much as she ever had. She was enjoying almost none of it, and she’d lost the handful of relationships that used to be built by the solving itself.

The team can tell the difference, even when they can’t name it. They can feel when a decision came from someone who sat with the discomfort of not knowing and brought other people into it, and when it came from someone who ran it past a system that agrees with almost anything stated with confidence. The outcome might be identical. The trust isn’t built the same way.

And the cost doesn’t stop with her. Every problem she used to think through with someone was also a problem that someone got to think through. That’s how a VP learns what a hard people decision actually involves, by being in one while it’s still unresolved, watching someone more experienced be unsure out loud and work it out anyway. When that gets routed through AI instead, the decision still gets made, often well. But it gets made in private. The person who would have grown by being asked doesn’t get asked, and doesn’t know what they missed.

Do that across an organization for a year and you get a strange result. The AI is getting stronger. The humans are only getting faster.

Deciding to walk down the hall and ask the one person who’d actually know isn’t more efficient. It’s a statement about what kind of leader you’re going to be now that more of the work can happen without you.

So before you hand the next one off, ask what gets lost. Not time. Not effort. More importantly, trust, presence, the two or three people who used to sharpen you before anyone else saw the work. If the answer is nothing, hand it off without guilt. If it isn’t, that’s the one you keep.

My client didn’t need to get sharper. She needed to start bothering people again.