A Stay Human Summer for You and Your Kids

AI is absorbing the work that comes apart into pieces, and the work that stays human is the work that won't—work that needs someone to be there, that braids many fields at once, that turns on judgment and taste and someone willing to own the result. You can read the version for you in Chapter 9 of Stay Human: The Work That Remains Yours

A Different Way to Think About Thinking

There's a bit of science worth knowing before you plan anything. We used to think thinking happened in the head, in words. Turns out it mostly doesn't. You think with your hands, and with the space around you. An architect doesn't draw a plan she already has in her head—she works it out by drawing, and she sees things in her own sketch she didn't know she knew. Stop someone's hands while they talk and their reasoning gets worse.

There's an inside version of this too, called interoception. It's the gut feel, the sense that something's off before you could tell anyone why. It's a real sense, and it's a big part of where judgment comes from. None of it runs on words.

And here's why that matters for your kid and AI. A machine can only work on thinking that's already been turned into words and symbols. That whole wordless layer—the hands, the body, the gut, the feel for a situation—is the part it can't get to, because that part was never in a form you could hand over. So the thing you're growing this summer is that layer.

Most advice about kids and AI is some version of "use it wisely," which is worth exactly nothing. Here is what the research actually implies, stated so you can act on it.

Build the skills that turn into judgment—and know that most of them are the ones we've called soft. When we measured AI's real capability against what jobs require, the usual advice inverted. The "hard" technical skills—coding, math, analysis—are precisely where the machine is furthest ahead. Skills such as reading a situation, holding a group through disagreement, earning trust, knowing when something is good—are where the human lead is widest, because they need a person to be there, a live social read, and years of practice that was never written down anywhere a model could learn it. These were never soft skills even though we called them that. They're the higher skills, and they're the ones to build on purpose:

• Holding a position without caving or bulldozing. Put your kid in something where they have to disagree with a person they like and stay in the relationship through it—and can't just quit when it's uncomfortable. A team that has an argument, for example. The capacity to stay in a disagreement long enough to find what's true is the exact thing a model can't do, because nobody inside it is accountable to anyone.
• Reading what isn't said. The shift at the check-in desk, the summer minding younger kids, the volunteer role with the public—anything where they have to notice that the quiet one is about to cry or the customer is angry about something other than what they're saying. This is the "active listening" a machine can score high on and cannot actually do, because there's no one in there building a model of another mind.
• Earning trust with the same people over time. The thing they keep showing up for—the two-year volunteer gig, not the one-off. A reputation built by being reliable to a real community is the least scrapeable asset there is.

Teach the author's posture, not the user's. There is a categorical gap between a kid who says "AI told me" and one who says "I used AI to test whether my argument held up." The first has handed over their mind. The second has recruited AI into it. You build the second by making arguing with the machine the normal mode in your house. When your kid brings you an AI answer, the question is never "did it work"—it's "how do you know it's right, and where would it break?" You're training the single most valuable professional reflex there is: treating output as a draft to interrogate, never a verdict to accept.

Use AI to reach the edge of a subject, then push past it. The future superpower: have them ask AI to lay out everything a field already knows—the settled middle of a subject—and then immediately argue with it. What's contested? What did you leave out? Where would a real expert disagree? Used this way, the tool compresses the years it normally takes to master the known and drops a curious kid at the genuine frontier, thinking muscles intact, because they did the wrestling alongside it instead of handing it over. The kid who argues with the machine reaches the edge years faster. The kid who just takes its answers never finds the edge at all.

Now the plan.