Over the past few months I've spent more time with AI tools than with Figma. Not out of conviction, but out of curiosity. I wanted to know what was actually there — behind the demos, the threads, the breathless takes about how design is supposedly reinventing itself right now.

What I learned doesn't quite fit into the hype cycle. It also isn't the counter-take you can usefully post as a designer right now. It's something third, with traces of both: the tools really are good. And they don't solve the problem they claim to solve.

The problem isn't the technology

The models are impressive. I mean that without irony. What Claude, ChatGPT, Midjourney and their relatives produce today was science fiction three years ago. Technically, there's nothing left to make fun of.

What increasingly bothers me isn't the technology. It's the attitude that's baked into most of these tools.

Every AI design tool on the market has an opinion about how I should be working. Not in the sense of defaults, but in the sense of workflow. You describe your project, the tool suggests a structure, walks you through steps, produces a result. You end up with something that looks like design but didn't feel like design.

That's the point where I start to bristle. Because design, for me, has never been linear.

Assistant or supervisor

The difference between a good and a bad AI tool in a creative context has, for me, become very simple to name: does it see itself as an assistant or as a supervisor?

An assistant waits for what I need. It knows my context, my preferences, the half-formed thought I had two hours ago. It produces when I want to produce. It holds back when I want to think. It accepts that I sometimes spend three hours running in one direction and then discard it, without that being a mistake.

A supervisor has a method. It wants to see that method applied. It asks you about your goals before you have any. It demands inputs before you have material. It turns every project into a workflow — and every workflow into a billing model.

The vast majority of AI design tools available today are supervisors. They package this as structure, as best practice, sometimes as "empowerment." But what they actually do is impose their idea of design on you — complete with a timer and a token budget. And because design is iterative, because it lives off discarding, every one of those iterations feels like a cost. That's not an environment in which good work gets made.

The tool that wins won't be the most powerful one. It will be the one that fits into existing processes instead of forcing new ones.

The leverage comes from somewhere else

The second point is the one I miss most in the hype debate: AI gives experienced people a lever they didn't have before — and it widens the gap to those without that experience.

That's an uncomfortable sentence because it contradicts the democratization narrative. "Now anyone can be a designer." In theory, yes. In practice, no.

The gap between "using AI tools" and "knowing what to do with them" is bigger than it looks from the outside. What I can produce in a good exploration over an afternoon looks like magic to someone without experience. What it actually is: fifteen years of evaluating, discarding, starting over — now with a faster tool.

The ability to explore quickly assumes you've already seen a lot and can judge it. Taste, judgment, the feel for when something is done and when it isn't — that's the part that doesn't get automated. The part that comes from work, not from subscriptions.

For experienced small teams, that's one of the best pieces of news of the past few years. For everyone else, it's a hint that the real investment is still the same one it always was: in your own judgment.

Freedom isn't a nice-to-have

The thought I want to close on is the most important one. It sounds soft, but it isn't.

Design needs freedom in the process. Not as a luxury, not as a privilege, but as a basic condition. Without the ability to put three variations side by side and throw two of them away, without the right to declare an idea bad, without the room to spend half an hour going in the wrong direction — design doesn't happen. Output happens.

Tools that fence in that freedom — with token limits, with paywalls, with workflows that have to be walked through linearly — haven't built a design tool. They've built a time-tracking system with a graphics department attached. That's an entirely different thing.

The real promise of AI in creative work isn't that it produces more. It's that it lowers the cost of discarding. That it lets you pursue five ideas where only one used to have time. That it makes exploration economical again.

Today's tools rarely deliver on that promise. But it's the promise I measure them against. And it is, I think, the standard we should be working toward as a profession — not faster generation, but better conditions for the work we're already doing.

What remains

AI in everyday design work is neither the revolution nor the apocalypse. It's a tool that fits some parts of the work extremely well and others not at all. Knowing which part is which is the real competence of the coming years.

The question isn't whether designers will be replaced by AI. The question is which designers will recognize where AI actually helps them — and where it only looks like it does. Telling those two apart requires exactly the thing that can't be prompted: experience.