I don't know enough about AI yet

I hear this constantly, and I said a version of it myself for most of last year.

It sounds like modesty. It's really a measurement, and it's measuring the wrong thing. It assumes there's a body of knowledge about AI you're behind on, and that the difference between you and someone making useful things with it is that they read more.

That's not the difference.

Before any of this, I ran customer experience for four small hotels in the Yucatán. Ten departments, fifty-odd people. I could tell you, without looking anything up, which complaints meant somebody was going to leave a bad review and which ones were just someone having a long day. Nobody ever asked me to write that down. There was no reason to. It lived in my head and it came out in decisions, one at a time, for two years.

Everyone's got a version of it. Which regulars you can seat late without asking anyone. Who on your team isn't ready to be in front of a client yet, and why you've never had to explain that to a single person. The one invoice question that means trouble, as against the four that don't.

That's the stuff that turns out to matter now, and almost nobody counts it as knowing anything. It isn't expertise in the impressive sense. It's the ordinary operating knowledge you pick up by being responsible for something, and it's invisible to you precisely because you've never had to say it out loud.

So I don't think the question is whether you know enough about AI. You won't get further by learning more about the tools. You'll get further by getting specific about something you already understand, which is a different job and a much shorter one.

Two things before this turns into a pep talk.

The first is that getting specific is the actual work, not the warm-up before it. Which is an annoying thing to hear, because it produces nothing you can show anyone. An evening spent writing down how something really runs leaves you with a text file. An evening spent reading about AI leaves you feeling like you learned something. One of those is progress and the other one feels like it.

It's harder than it sounds for a reason that took me a while to work out. You can't sit down and list what you know. I've tried. The material worth having is the material you stopped noticing you had, and that's exactly what doesn't come when you ask yourself for it directly. It's automatic by now. That's what makes it valuable and it's also what makes it invisible.

What works instead is catching yourself correcting something. You read an answer and something in you goes no, that's not how that works. That reaction is the knowledge arriving. Most people have it and move straight on, because the correction takes two seconds and doesn't feel like information. Write it down instead of just making it. Do that four or five times in a week and you'll have something better than anything you'd have produced by sitting down and trying to be thorough.

It also isn't documentation, which is the thing people hear and then dread. Nobody else has to read it. It doesn't have to be organised. It will be slightly embarrassing if anyone does read it, because it'll say things like the Tuesday report is always wrong in the same place and I don't know why. That's fine. Documentation is written so another person can follow it. This is written so a machine stops guessing, which is a much lower bar and a much shorter evening.

The second thing is that it fails, in specific ways, and they're worth knowing about before you spend that evening. This week gets into them. A week that only tells you something works isn't worth reading.

What I'd watch for, if you read nothing else: notice the next time you correct something an AI wrote about your own field. Whatever you knew that it didn't is the thing I'm talking about.