Taste is not a mystery

There is a line going around that taste is the differentiator now. That when anyone can generate anything, the only edge left is taste. The people saying it are right that it matters and wrong about nearly everything else, because most of them cannot tell you what taste actually is. They treat it like something you either have or you don’t. I want to take it apart, because taking it apart is the skill, and it is what changes how you work with these tools.

I recently designed and built the launch site for Omaru by directing an agent. I set the direction, it generated, I cut and corrected until it was what I wanted. Working that way puts all the weight on the one part everyone mystifies, so that is where I will start.

Taste comes apart

Start at the bottom. There is a layer of taste that is just the language of visuals. Color, composition, balance, contrast, hierarchy. The same vocabulary a painter, a photographer, a cinematographer spends a career inside. These are not opinions. They are the building blocks everything visual is built from.

Above that sits coherence. Not whether one screen looks good, but whether all of it feels like one hand made it. A photographer has it across a series. A cinematographer has it across films. In a product it runs at every scale, the app and the landing page as two paintings in one show, or one feature against the next. This is the part people find hardest to name, so they just call it taste.

A model is extremely good at both of these, the moment you can name the concepts. The people saying craft is the differentiator now have usually never really done it themselves, because craft, the foundations and the coherence, is exactly the part AI is already good at. If you know what to ask for, you get it.

The layer everyone gets wrong

Then there is the layer that actually decides whether the work is any good, and it has nothing to do with looking nice. Something can be beautiful and still not work. What makes work matter is relevance. Whether it communicates the idea, is easy to understand, and fits the culture of the person you made it for.

I used to think this was the part a model could not reach. I was wrong. All of it is already in there. Every Swiss poster, every album cover, every anime frame, every brutalist facade. Ask it for the visual language of a 1970s Braun manual, then for a Y2K rave flyer, and it gives you both, because it has read the whole library. You just have to tell it which part you want.

So the edge was never a mystical feel for the moment. It is the map and the vocabulary. Knowing which visual language belongs to the person in front of you, and being able to say its name. And that map is old. What we call design in tech is graphics and visuals, and the older crafts, poster design, album art, film, photography, comics, architecture, were fluent in all of it long before there was a screen to put it on. Tech design is the newest of all of them, and the people who are best at it usually know the older ones well.

You can get there

So if AI can handle the craft, what is left for you is the deciding, and that turns out to be the whole game. Design has always been a loop, you build a rough version to see it, judge it, adjust, and build again. Building was the slow part, so you learned to judge in your head first and kill the weak ideas before wasting a day on them. That fast internal cut looked like the talent. Some of it was. But a lot of it only mattered because building was slow.

Now building costs almost nothing. You do not have to ration it or do all the judging in your head. You build the real versions and judge those instead, and the workaround falls away. What is left underneath it is the judgment itself, knowing what to keep. That is taste, out in the open. Every time you look at a version and say no, wrong, or yes, that one, push it, you are running the foundations and the coherence and the relevance over it and deciding.

It only works if you can say why. Being able to name why a version fails is what lets you fix the next one. If all you can say is that it feels off, you just make the same mistakes faster.

This is why taste is suddenly the thing everyone talks about. Once AI gives everyone the same speed, the only thing left to separate the work is the deciding. And deciding well is not a gift you were born with, it is just how well you understand your craft, which is something anyone can go and build.

It starts with looking

Taste is not a gift and it is not a mystery. It is a language, foundations, coherence, relevance, and like any language you can learn it, name it, and use it to tell a very fast, very literal AI exactly what you want. And learning it is not a grind. It means paying more attention to the things you find beautiful, and working out why they are beautiful. If you want better work out of these tools, that is where it starts. And hey, that doesn’t sound horrible, does it?