Where I See the Line Between Delegating and Forgetting

This morning I dictated a voice note to my AI, ran it through a workflow I built, and published the result. The note was about people handing their thinking over to AI. So yes, I get the joke, and I let it stand instead of rewriting it, because the joke is the whole argument.

Here's what got me thinking about it in the first place.

There's a study from 2000 where researchers scanned the brains of London taxi drivers. They found the hippocampus, the part of the brain that handles spatial memory, was bigger than average in those drivers. That makes sense on its own: London is a maze, and those drivers spent years learning it by hand, street by street, before anyone could look up a route on a phone. The part that stuck with me is what happened after they retired. It shrank back to normal. The brain built that capacity because the job demanded it every day, and when the demand stopped, so did the capacity.

There's a related finding from around 2011 out of Columbia. When people know a fact is stored somewhere they can get back to, they remember where it lives instead of what it says. The filing location survives. The content doesn't. Your brain makes that trade because it's the cheaper one, and it doesn't ask you first.

I want to be clear that I'm not writing this as someone backing away from any of it. I'm as pro-AI as anyone you'll find. I run seven sites and most of a business on it, and I'm not going back to doing it the old way. That isn't the concern.

The concern is that I watch myself write prompts now, and they're getting lazier. Less time spent working out what I actually want before I ask for it, more just handing it over and seeing what comes back. I used to think a request through, decide what good looked like, then write the prompt. Now there are days where the prompt is the first draft of my thinking instead of the last, and the difference is easy to miss from the inside because the output still looks fine.

Social media was a dopamine problem. Everyone understands that one by now, and the whole conversation about screen time was really a conversation about a slot machine in your pocket. This is a different mechanism. It's a delegation problem, and I think it goes deeper, because delegation is a thing we've all agreed is good.

And mostly it is. Everyone who becomes a CEO loses touch with the work eventually. The person who founded a plumbing company stops going out on calls, then stops quoting jobs, then stops knowing what a water heater costs this year. That's normal and it's fine, because delegation is what keeps the engine moving once the engine is bigger than one person. It's also recoverable. That founder can walk out to a truck and follow along, because they know what the work is even if they've gotten slow at it. The skill is dormant, not missing.

Here's the split I keep coming back to. Delegating work you know how to do is one thing. You can check it, you can tell when it's wrong, and you can take it back if you have to. Delegating work you never learned in the first place is something else entirely, because there's no version of you that could evaluate what came back. You get an answer and the only thing you can do with it is accept it.

That second kind is what's new. A founder who stops quoting jobs still quoted jobs for years. Someone whose first thousand pieces of writing were generated has no earlier version of themselves to fall back to. The capacity was never built, so there's nothing to shrink.

I don't think the answer is doing things by hand to prove you can. That's just nostalgia with extra steps, and it won't survive contact with an actual deadline. But delegating where you have no business delegating, or worse, where you have no idea what you're even delegating, that can't be good. If you can't describe what the task involves, you can't tell whether it was done well. You're not managing at that point. You're just hoping.

So the thing I've started doing is small. Before I hand something over, I check whether I could describe what a good version looks like. Not do it, just describe it. If I can't, that's the signal that I'm delegating something I never learned, and I slow down and work out what I'm actually asking for before I ask.