Outsourcing the Thinking Part

Been thinking about a couple of things that landed on my radar this week, and they’re really the same story told twice. A US school district rolled out Google Gemini for its students, then a few weeks later banned AI outright across every grade. And separately, there’s a study doing the rounds claiming that just ten minutes of AI use measurably erodes your ability to persist at hard things. Put those next to each other and you’ve got a fairly blunt picture of where we’re at.

I work in DevOps. I use AI tools most days, sometimes to scaffold something tedious, sometimes to sanity-check an approach before I commit to it. And I’ll say this plainly, because I don’t think there’s much point pretending otherwise: I’ve noticed the same pull in myself that the commenters on that study were describing. There’s a particular kind of friction that used to be the job, working through a problem that doesn’t want to resolve, sitting with the discomfort of not knowing, and that friction is exactly what AI removes. The problem is that friction was never just an obstacle. It was the thing doing the training. Remove it enough times and the muscle that handles hard problems quietly atrophies, and you don’t notice until the day you actually need it and it’s not there.

One comment on the school district thread stuck with me more than the rest: giving kids AI in the classroom is like handing them a golf cart when it’s time to run laps. The point was never the laps themselves. Nobody cares if you can recite times tables as an adult or parse a quadratic equation on sight. The point was always what doing the hard, boring, repetitive thing does to the brain doing it. We don’t make kids deadlift a barbell because they’ll need to move barbells later in life. We do it because the strength transfers to everything else. Outsource the rep and you don’t get a stronger kid who’s also faster now. You just get a kid who never got strong.

My daughter’s in high school, and the AI question comes up constantly in our house, usually at the dinner table, usually started by her complaining that a friend got away with using it for an assignment and she didn’t. I don’t have a tidy answer for her. I genuinely don’t know where the line sits between “legitimate tool, like a calculator” and “doing the thinking for you, like paying someone to run your laps.” The comparison to calculators came up a lot in that thread too, and one tutor made a point I keep coming back to: a calculator breaks down the moment you introduce an unknown, an X, and that’s exactly when you find out whether a kid actually understands what’s happening underneath the arithmetic. AI doesn’t break down like that. It just keeps producing a confident, plausible answer regardless of whether the user understood a word of it. That’s a much bigger problem than a calculator ever was, because the gap it creates is invisible for far longer.

Here’s the part I can’t resolve and won’t pretend to: I think banning AI in schools outright is almost certainly the wrong call in the long run, because these kids are going to live and work alongside this technology whether we like it or not, and pretending otherwise doesn’t prepare them for anything. But I also think the uncritical, full-send rollout that preceded the ban was just as wrong, maybe worse, because it handed over the cognitive weight room to a chatbot before anyone had worked out what the kids were supposed to be building in the first place. Both reactions feel like overcorrections to the same underlying failure, which is that nobody did the actual work of figuring out what AI is for in a classroom before the decision got made one way or the other.

What I keep landing on, for myself at least, is something closer to the way I try to think about Elden Ring, of all things. The whole appeal of that game is that it refuses to make things easy for you, and the difficulty is the point, not an obstacle to the point. I don’t want a mod that walks me past the hard boss, because then what did I actually get out of the forty hours. I think there’s a version of using AI that respects that same principle: let it handle the genuinely mechanical stuff, the boilerplate, the formatting, the thing that was never teaching me anything to begin with, and keep the actual hard thinking for myself, because that’s the part that was building something.

I don’t think we’ve worked out as a society what that line is yet, for kids or for forty-five-year-olds in IT jobs. I’m not sure the AI companies have much incentive to help us find it either, given that the more indispensable the tool feels, the better it is for them. So we’re left to work it out ourselves, one dinner table argument and one quietly abandoned train of thought at a time. Not a neat ending, I know. But I’d rather sit in that than pretend either the ban or the rollout had it figured out.