The Taste Differential

I was in a conversation with a designer yesterday about what the job actually is now. We were looking at an iOS app I built in Claude Code, no Figma files, just shipped. He said “That’s crazy. You got all this just from Claude Code.” I told him I’d taken it to 85. The harder part is taking it to 100.¶

Then I said: “You don’t need people who just think in UX anymore. You need people with good taste. Claude does a lot of that work, and then you need people to say ‘that sucks, change that.’”¶

AI didn’t commoditize the whole design job. It commoditized the middle. The flows, the spec, the systems retrofit, the polished mock. The part most of us got hired on. A solo designer with the current tools produces that whole stack in a week, sometimes a day.¶

The ends still matter. UX thinking and product thinking still matter, probably more than they did, because they’re what initiates the work now. You can’t prompt Claude to build something good if you don’t know what good looks like, who it’s for, or what the actual problem is. That’s the front end of the work, and it still has to come from a person.¶

What changed is that the thinking no longer plays out through deliverables. You don’t draw a journey map to know the journey. You hold it in your head, decide what matters, and start building. Claude takes the middle. Your judgment closes it. That’s the new shape of the work. Thinking at the front, taste at the back, automation in between.¶

The back end is where the differential lives. AI doesn’t know which of the seven equally-correct outputs is the one your product should actually ship. That call is the job.¶

Most uses of the word taste are cheap. I don’t mean the LinkedIn-fire-emoji version where everyone has an opinion and AI confirms theirs back. I mean the eye that looks at something working and sees what’s wrong before it can explain why. The conviction to be right about that more often than not. The editorial discipline to cut the thing you made when it isn’t earning its place. That capacity is what AI can’t compress, and most of us spent the last decade hiding it behind process because process was easier to defend in a job description.¶

Here’s where I was wrong. Six months ago I’d have told you Figma was being made redundant. I’m back in it. Not as the place I do the work, but as the place I do the part of the work that needs a different kind of attention. The 85-to-100 push pulls me back into the canvas, alone, for an afternoon, on a small visual decision the canvas is still better for. It’s not a regression. It’s the move taste makes when it needs that tool.¶

That’s the new process. The tool I reach for changes decision by decision. The old answer was that the tools dictate the workflow. The new answer is that the taste dictates the tool, and the workflow rearranges itself around what the call needs. AI gets you a 70. Process gets you to 85. The remaining 15 is judgment, and judgment doesn’t come from a framework.¶

The designers I see pulling away are doing less of the middle and more of the bookends. They’ve offloaded everything the tools handle and put the time back into the front (framing the problem, setting the constraint) and the back (opening the output, picking the tool, shipping the fix). The leverage is in the speed and accuracy of the diff between what came back and what should have. That’s the taste differential, and it’s the only part of the job that compounds.¶

The move isn’t to defend the middle. It’s to invest in both ends: the thinking that initiates the work and the eye that closes it. Look at more good work. Look at more bad work until you can name exactly what makes it bad. Ship things end-to-end until you can look at a working product and know what to change.¶

AI absorbed the middle. Exist on the edges.¶