Taste Is the Skill AI Can't Give You

Last month I shipped a landing page that an AI generated in about four minutes. Clean layout, responsive, on-brand color, sensible spacing. It passed every check I could name. Two days later I quietly killed it and rebuilt the hero myself, because the original was boring and I was the only person on the thread who could say why.

That "why" is the whole job now.

Here is the belief I want to push against: that AI is coming for taste the way it came for production speed. It isn't, and the data is already saying so. Figma's 2025 AI report, built on a survey of around 2,500 designers and developers, found that 78% say AI makes their workflows more efficient, but only 58% say it improves the quality of their work. Speed went up. Quality mostly didn't. The gap between those two numbers is where taste lives, and it is getting wider, not narrower.

We even named the fallout. By the end of last year the phrase "AI slop" had gone fully mainstream, enough that Euronews ran a December 2025 piece calling 2025 "the year AI slop went mainstream." Slop is not a technical failure. Every piece of it is grammatically fine, correctly rendered, technically competent. It is bad in the one way machines can't yet catch on their own: it has nothing to say, and no one making it noticed.

What taste actually is (it's not a vibe)

People talk about taste like it's a mood, or something you either have or don't. It isn't. Taste is a trained judgment about what's good, aimed at a specific context, that you can act on under pressure. It has three parts that AI genuinely struggles with.

First, discernment: seeing the difference between two options that are both "fine." AI will happily give you a hundred acceptable homepages. It will not tell you that number sixty-three is the only one with a point of view, because it doesn't have a point of view to measure against. You do, or you're supposed to.

Second, restraint: knowing what to leave out. Generation is additive by nature. Ask for more and you get more. Taste is almost always subtractive, the decision to remove the clever animation, the third font, the feature everyone wants and no one needs. I've argued before that the best AI work is the stuff that resists the gimmick, which is really a story about thoughtful AI in product design over flashy output. Restraint is a taste muscle, and the tools pull the other way.

Third, intent: knowing what the thing is for and who it's for, at a level deeper than the prompt. A model optimizes for the average of everything it has seen. Taste optimizes for this user, this brand, this moment, this one decision that makes the product feel like someone was home.

Why the AI era raises the price of taste

The reflex is to assume that when a machine can produce a competent screen in nine seconds, the human's value drops. The opposite is happening. When competent is free and infinite, competent stops being a differentiator. The scarce thing becomes the judgment to pick, cut, and point the output somewhere that matters.

This is the part of what vibe coding actually means for designers that gets missed in the hype. Describing an app and watching it appear doesn't make you an engineer, and it doesn't make the AI a designer. It moves your job up the stack, from making the thing to deciding whether the thing is any good and knowing what to change. That decision was always the expensive part. Now it's the only part that's scarce.

I felt this most on editorial work, long before the current wave. Building the El País multimedia storytelling portfolio, the entire value was judgment calls a brief could never contain: which image earns a full bleed, where the silence goes, when to break the grid and when breaking it is just showing off. No generator produces that, because the right answer isn't in the training data. It's in the context, and context is exactly what taste is tuned to.

The uncomfortable part

Here's what nobody selling an AI tool wants to say out loud: taste is built the slow way, and AI quietly removes the reps that build it.

You develop discernment by making bad things, noticing they're bad, and figuring out why. By sitting in the gap between what you pictured and what you made. If AI does the first ninety percent every time, a whole generation of designers may never sit in that gap long enough to build the judgment to evaluate the output they're now responsible for approving. The tool that makes you faster can also keep you a beginner, confidently.

So I use AI constantly, and I treat the output as a draft, never a verdict. It generates; I decide. It covers the competent middle; I spend my attention on the edges, the cuts, the one choice that gives a screen a spine. The machine made the page in four minutes. Deciding it wasn't good enough, and knowing what "good enough" even meant, took fifteen years. That part isn't for sale, and it isn't in the prompt.

What's the last thing you made that was technically perfect and still not good? The answer to that is your taste talking. Keep listening to it.