I went to a local AI meetup last night. Good evening, sharper crowd than these usually are. The topic was using AI to throw together quick prototypes, and the most interesting person in the room was a founder who has stopped briefing designers altogether. He builds the interfaces for his customers himself now, in Claude Design, and he is evangelical about the speed. No briefs. No back and forth. He doesn't have to explain himself four times over any more, and that is the part he loves.
Here is the part I want to be honest about, because the rest of this only works if I am. He is right. The things he showed were faster to produce than anything he would have managed in Figma, and they looked better than anything he would have managed in Figma. If you had told him to learn the tools and do it by hand, he would have produced something worse and spent a fortnight doing it. The velocity is real, and the output cleared his own bar with room to spare.
So why did I spend the train home thinking about it.
Because nobody in that room, including him, could tell me whether the designs were any good. They could tell me the designs were finished. That is a different claim, and the gap between the two is the whole subject of this piece.
Why you can't tell
To see why that gap matters, go back about ten years, to the moment UX became a desirable thing to have on a business card. What happened next was predictable. Graphic designers and web designers, more or less overnight, became UX designers. Same person, same skills, same output, new title, and a new salary expectation to match. For anyone without a trained eye, and that is almost every buyer, this simply became what UX was. The reference point for a generation of founders got set by people who had relabelled themselves and changed nothing about how they worked.
So when I say most people have never worked with a UX professional, I am not being snobbish. I mean it literally. They worked with a screen designer who called themselves a UX designer, and they have no reason to know the difference, because the difference was never made visible to them. The research that is meant to sit underneath the pixels was usually absent, and nobody was checking.
Strip away the title and the thing underneath is simple. The interface is the icing. The cake is the harder question, the one that takes real work to answer: do your customers actually want the thing you are about to build, in the shape you are about to build it. Nice icing on a cake nobody wants to eat is still a wasted cake.
What the tools actually changed
Now put AI into that picture. The founder at the meetup has not solved the cake problem. He has automated the icing, and automated it well. The tools are very good at producing interfaces that look considered, that carry the visual grammar of something a competent professional made. What they cannot do is tell him whether anyone wants what he has built. They produce something that looks like sugar and does not taste like it. He can't tell the difference, for the same reason his customers can't, and for the same reason the market couldn't for the past decade: the finish was always the part that was easy to see and easy to fake.
If you want proof the industry already knew this, look at Dribbble. For years the UX/UI section was full of beautiful screens no user had ever touched and no researcher had ever pressure-tested. Everyone serious understood the deal. This was portfolio theatre, pretty pictures to win clients, not products that had survived contact with reality. We tolerated it because it stayed in the gallery. What the founder is doing, without realising it, is shipping Dribbble. The exact thing the profession waved through as decoration is now the default way to build the real thing, aimed straight at paying customers, with the one step that used to catch it quietly deleted.
How to actually decide
None of which tells you, the person deciding whether to use these tools, what to do on a Tuesday. So here is the part that is useful rather than just true.
The question is never whether the design looks good. You can't judge that from a screenshot and neither can I. The question is what a wrong decision costs you and when you find out. Sort your work that way and it falls into three piles.
The first pile is low stakes and easily reversed. Internal tools. A landing page for a campaign that runs three weeks. A prototype you are using to think with. If the design is mediocre you will change it next month and lose nothing. Here you should use the AI tools, hire nobody, and feel no guilt about it. A professional adds nothing you should pay for. I want to be plain about this, because the rest of the argument is worthless if I am secretly routing every example back to “and that is why you need someone like me.” You don't, not here.
The second pile is the dangerous one, and it is dangerous precisely because it looks like the first. The decision is invisible now and expensive later, and you will almost never trace the cost back to its source. The onboarding flow that quietly suppresses how many people ever reach value. The information architecture that is fine at forty items and incoherent at four hundred. The permission model that works for one user and falls apart for a team. You will experience these as something else entirely. High churn. Rising support tickets. A feature that didn't land, and no clear reason why. You will go looking for the cause everywhere except the interface decision you made eighteen months earlier, because by then it doesn't look like a decision. It looks like the furniture. AI is at its most dangerous here, because the polish removes the friction that used to make someone stop and ask whether this was right. The mess used to be the warning. There is no mess now.
The third pile is where the decision is the product. A regulated workflow where getting it wrong is a liability rather than an inconvenience. A core interaction that is the entire reason a customer chooses you over a competitor. Here the judgement is the thing of value, and the AI is a drafting tool in the hands of someone who already holds that judgement, not a substitute for having it.
If you want to keep the whole thing to three questions, ask these before you ship the AI version of anything. If this decision is wrong, when do I find out. What will it cost to reverse by the time I do. And can I, honestly, tell the difference between good and merely finished here. Soon, cheaply, and yes: ship it, and don't apologise. Late, expensively, and no: that no is the entire reason the role you are tempted to skip exists.
For ten years the icing and the cake looked the same from the outside, and plenty of people made a living on the fact that you couldn't tell them apart. The AI tools did not create that problem. They removed the cover for it. They made the icing free, which leaves the judgement about the cake as the only thing worth paying for, and that was always the rare part, title or no title. The masks were going to come off eventually. It turns out the thing that pulled them off was a founder at a meetup who is delighted, and entirely unaware of which half of the job he has automated. ✳