There's a version of the AI story that everyone is telling right now. Productivity up. Output faster. Teams doing more with less. It's true. I've seen it firsthand in my own work, in my own team. The ceiling is genuinely higher than it was two years ago.

But there's another version nobody is talking about honestly yet. And it starts with a simple question I haven't been able to shake: what happens when the tool is wrong and nobody on the team catches it?

Not a hypothetical. A real failure mode I think about more the deeper we go into this.

Here's the thing about AI tools that we don't say out loud enough. They're confidently wrong. Not confused, not uncertain, not flagging their own gaps. Just wrong, in complete sentences, with the same tone they use when they're right. And if you don't already know enough to catch the mistake, you won't. The output looks like an answer. It feels like an answer. It ships like an answer.

The problem isn't the tool. The problem is what happens to judgment when people stop exercising it.

I've watched this happen in small ways. A junior engineer takes the AI's suggestion without questioning whether it fits the actual context. A document gets generated and reviewed for formatting rather than for whether the logic holds. The tool becomes the thinking, and the person becomes the hand that presses submit.

It varies by person. Some people use AI and get sharper. They push back on it, interrogate it, use it to go faster on the parts they already understand. Others just accept whatever comes out. And the scary part is that both kinds of output can look identical on the surface.

The ceiling question is what keeps me up. I realized early that AI makes the capable person dramatically more capable. That felt like good news. It still is, for me personally. But zoom out a little and it creates a different kind of problem.

The gap between someone who thinks well and someone who doesn't is getting wider, faster. Because AI amplifies whatever you bring to it. If you bring judgment, you get leverage. If you bring none, you get volume. And volume that looks like output is one of the harder things to manage, because it's not obviously wrong until something breaks.

I'm building a team. I'm also deploying AI deeper into how that team works. Those two projects are in tension in ways I didn't fully anticipate. The goal was always to get to a place where the team doesn't need me for every decision. But every AI capability I add makes that harder in a specific way: it removes some of the friction where the learning used to live.

Before, a junior engineer had to figure something out. The struggle was annoying but it built something. Now they can skip straight to an answer. Faster, yes. But what did they just not learn?

I don't have a clean answer to this. I want to be upfront about that.

What I've landed on for now is something like: the job of a leader in an AI-enabled team isn't to teach people how to use the tools. It's to keep insisting that the tools aren't the thinking. That the output is not the work. That knowing whether something is right matters more than knowing how to produce it quickly.

That's harder to instill than a workflow. It requires people who are genuinely curious, not just capable of following instructions. And it means I now care more about how someone reasons through a problem than how fast they can produce something.

The question I used to ask in hiring was roughly: can this person build? The question I ask now is different. Can this person tell when something is wrong? Do they slow down when something doesn't feel right, or do they just ship it because the tool said so?

Those are different people. And the second kind is rarer than I expected.

So here's where I've landed, for now.

AI is making the ceiling higher. That's real. But it's also making the floor easier to fake. And in a small team where I can't review everything, where I'm trying to build people who can own things independently, the thing I'm most afraid of isn't the tool being wrong. It's the possibility that we've built a team that can produce without thinking, and we won't find out until something important breaks.

I'm still figuring out how to solve that. I suspect most people building teams right now are in the same place, whether they're saying it or not.

The tool isn't the problem. It never was. The question is what we do with the judgment that the tool can't replace, and whether we're doing enough to make sure it doesn't quietly disappear.