The 2–7 problem

Emmett Shine said something a while back that I haven’t been able to shake.

He was talking about AI output — not in the abstract, but on a scale. Picture a one-to-ten grade for quality. A 10 is the thing you remember for years. A 1 is the thing that makes you stop scrolling because it’s so off it’s interesting. A 5 is competent. Forgettable. Fine.

Most things in the world live somewhere on that scale, and most of them sit around the middle.

His point: AI output never leaves the middle. It always lands somewhere between a 2 and a 7. Never a 1. Never an 8.

The line that stuck: AI is bad at making things that are bad.

The obvious half

The first half of that sentence is the part everyone agrees with. AI can’t make a 9 or a 10. You see it everywhere. The vibe-coded app that works but doesn’t feel like anything. The AI-generated article that’s competent in a way that makes you forget it the moment you close the tab. The image that’s technically correct and emotionally inert. None of it is bad. None of it is great either. It’s just there, taking up space, doing the job.

I’ve written about the missing ingredient — taste, judgment, point of view — and so has everyone else by now. That argument is settled. Fine.

It’s also the less interesting half.

The less obvious half

AI can’t make a 1 either.

It can’t make work that’s tonally off in the way only a person can be tonally off. It can’t make something accidentally brilliant. It can’t make something so specific and weird that you remember it three years later. It pattern-matches toward the median of acceptable, which means the floor is a 5.

That’s higher than most professional work has ever sat.

The ceiling argument is comforting. AI can’t replace the best. True, and irrelevant for most people. The floor argument is the one that actually matters. AI raises the median to a 5. That’s where the displacement happens. Anyone who was already at a 5 wasn’t safe before AI. They’re less safe now.

The irony

Here’s the part that’s been bothering me.

The only way anyone got to an 8 was by making a lot of 2s and 3s first. Not because being wrong is the lesson, but because you have to try a bunch of things before you know which ones are worth keeping. My early work wasn’t good. It wasn’t supposed to be. It was twenty versions of the same idea, and slowly I noticed that one of them had something the others didn’t. That’s the whole mechanism. Taste is the residue of trying a lot of things.

AI quietly removes that.

I notice it in my own work. I try fewer things now, because I get to something decent so quickly. The first output is a 5. Why would I do nine more? The friction that used to force me through the variations is gone. So I stop. And what I stop at is fine. It’s just not the eighth version, because there is no eighth version anymore.

This isn’t a junior designer problem. I’ve been doing this for nearly thirty years and I do it too.

What to do with that

I don’t think the answer is “use AI less.” That ship sailed and I’m on it.

The answer is closer to: notice when you’re stopping. The first decent output is now the thing to be suspicious of, not satisfied with. The work that used to happen between version one and version twenty is the work that built whatever taste you have. If you skip that part, you skip the part.

So: try more things on purpose. Make the version you’d normally throw away. Make the version that’s too much. Make the version that’s nothing like the brief. Most of them will be bad. That used to be a feature.

It still is.

The trap

The trap isn’t AI. The trap is that 7 is easier to reach than it’s ever been, and 7 feels like enough.

The middle has never been more crowded. The interesting work is on the edges — the 1s and the 9s — and only one of those edges is still accessible to a tool. The other one is accessible to you, if you remember how to get there.

The skill now might just be knowing what a 1 used to feel like. And being a little suspicious of anything that lands at a 5.