You can give someone my AI tools. You can't give them my triggers.

AI can dramatically increase the supply of content. I’m less convinced it increases the supply of things worth noticing.

You can give someone my AI tools.

You can’t give them my triggers.

I’ve been thinking about this because my field notes are getting increasingly weird.

I’ll watch a film and start thinking about ageing and work. I’ll play RimWorld and end up thinking about storytelling. I’ll see somebody complain about AI writing and start wondering whether we’ve confused writing with typing. Football becomes an experiment in evidence and judgement. A random LinkedIn comment sends me down a completely different path.

Some of it is cynical. Some of it is deliberately contrary. Some of it is probably a tangent that goes nowhere.

That’s increasingly the point.

Where AI actually joins

I want to be precise about this rather than romantic, because the honest version is less flattering than the version where the machine does nothing important.

AI is heavily involved in how I develop these ideas. I use it to challenge them, to connect them to things I’ve thought about before, to find holes in arguments, and eventually to help turn them into something publishable. Quite a lot of the words you’re reading arrived that way.

It just usually isn’t where the interesting bit starts.

The interesting bit starts with noticing something.

Why did that line in a film stick with me?

Why did that comment annoy me?

Why does something everybody seems to agree with feel slightly wrong?

Why does a computer game remind me of how organisations build knowledge?

AI can generate a hundred ideas for posts. That is not the same thing.

Because if the observation underneath a piece isn’t interesting, making the production process faster doesn’t solve very much. It produces the same thin thought, sooner, in better sentences.

What got cheaper, and what didn’t

I’ve argued elsewhere that the upstream work only became practical recently — that AI didn’t make research, synthesis and evidence valuable, it made them affordable.

This is the part of that argument I want to separate out, because they’re often collapsed together.

What got cheaper was developing an observation. Reading around it, connecting it to things you already know, testing whether it survives contact with the evidence, working out whether anybody has said it already.

What didn’t get cheaper was having one.

The noticing happens before any of the tooling is involved. It happens while you’re watching something, or irritated by something, or half-listening in a meeting. By the time a system is involved, the interesting decision has usually already been made.

The obvious objection, which comes from my own repository

Here is where I have to argue with myself, because I’ve published something that cuts against this.

I’ve written that my repository gives me back connections I didn’t explicitly put into it — that with enough accumulated context, retrieval starts producing links I hadn’t consciously made. That sounds a great deal like a machine noticing something.

I think the distinction holds, but it’s narrower than I’d like.

The system can connect things that are already in it. It cannot decide what goes in.

Every one of those connections was drawn between observations I had captured, because something made me stop and capture them. The film, the game, the annoying comment. The machine is very good at finding relationships across that material and genuinely surprises me. It has never once told me to go and watch something.

So the honest claim isn’t that AI can’t notice. It’s that noticing has two halves — what enters the system, and what gets connected once it’s there — and the tooling has become extremely good at the second while doing nothing at all about the first.

The other objection, which I like less

Triggers might be trainable.

I’ve described mine as though they’re a temperament, which is convenient for somebody who has them. But most of what I’m calling noticing is probably fifteen years of journalism turning into a reflex: the habit of asking why a thing is being said, who benefits, and what’s missing. That’s a craft skill, taught badly and slowly, but taught.

If that’s right, then “you can’t give them my triggers” is too strong. You can’t hand them over, but they can be built, and I’ve written before about how much harder that is for somebody starting now.

I’d rather leave the argument in that state than tidy it. The claim I’m confident about is only that attention is scarce and under-discussed — not that it’s innate.

What we’re getting wrong about AI and creativity

We focus enormously on the output. The writing. The image. The prompt. The model.

Almost none of the discussion is about what preceded it.

What did you notice?

Why did you notice it?

What did it connect to?

And did you have anything interesting to say about it?

AI can dramatically increase the supply of content.

I’m much less convinced it increases the supply of things worth noticing — and if it doesn’t, then the constraint on good work hasn’t moved at all. It’s just become more visible, because everything downstream of it got faster.