When Everyone Can Generate, Taste Is the Moat

Cheap production has happened before. Photography did it to painted portraits. Desktop publishing did it to typesetters. Each time, the same thing happened. People who sold their skill at making things lost ground. People who sold their judgement got a big payoff. We are living through a much faster version of that now.

First, what does moat mean? It is the thing that stops someone copying you. Here is the kitchen version. Anyone can buy the same ingredients, so the cook is the difference. A camera in every phone did not make everyone a photographer. A model can now build a decent landing page in twenty seconds. The contrast is readable, the spacing is sensible, the type looks fine. What it cannot do is know that this product should feel plain, not friendly. The buyer is a compliance officer. Their whole job is risk. To them, warmth reads like a sales pitch.

There is a mechanical reason a model cannot make that call. It works by predicting the most likely next thing, which is the middle of everything it was trained on. The middle is a sensible landing page for a generic company. It is the average of a million websites. Knowing that this buyer needs the opposite of the average means stepping outside the training data on purpose, and average is the one place a model always returns to. That gap, between the average and the right call for this exact person, is the whole moat.

Design is not just what it looks like and feels like. Design is how it works.

What taste actually is

Taste is not a fancy word for personal preference. It is a large library of remembered examples plus the ability to compare a new thing against that library fast. That is the whole mechanism. You have seen thousands of outcomes, good and bad, and a new option gets matched against them in a fraction of a second. The match feels like a gut call because you cannot watch the library working. But it is memory doing a search, not magic.

The clearest proof comes from chess. In 1973 William Chase and Herbert Simon showed masters a board for five seconds, then hid it and asked them to rebuild it from memory. On a real game position the masters were almost perfect. On a board of randomly scattered pieces they were no better than beginners. Their edge was never raw memory. It was a stored library of thousands of real positions, and instant matching against it. Take away the familiar patterns and the gift vanishes.

Design taste works the same way. A strong designer is not born with a better eye. They have looked hard at a huge number of interfaces, layouts, and typefaces, and filed away what worked and what did not. When they glance at a new screen, they run it against that file. This is also why taste is domain bound. A brilliant photographer has no taste in database schemas. The library is specific to the thing you have studied.

The library is built on purpose, not by soaking it up. Anders Ericsson spent his career studying experts and found the same pattern everywhere. The people who got great did deliberate practice, which means working at the edge of your ability, comparing your output to a standard, and closing the gap on purpose. Scrolling a feed adds nothing, because there is no comparison and no feedback. Redrawing a layout you admire until you can say why each choice was made adds a lot. You are not collecting images. You are labelling them.

That is also why taste fades in the exact world AI creates. If you make a hundred options and never live with the one you ship, you never feel the result. And the result is what labels the example so it can join the library. Volume with no consequence is not experience. It is just scrolling with extra steps.

Why pretty things feel easier to use

In 1995, Masaaki Kurosu and Kaori Kashimura tested twenty six ATM layouts with 252 people. They found something that still annoys engineers thirty years later. How pretty people found a layout predicted how easy they thought it was. And it predicted that far better than how easy it actually was. People forgive a beautiful thing its faults.

The lesson is not just make things pretty. It is that a good-looking choice is a trust signal, and trust drives every decision a user makes. When every rival can make a decent interface, decent stops setting you apart. What is left is the pile of small, careful choices a machine will not make for you.

How I try to keep judgement sharp

The library grows fastest when you label examples instead of just saving them. Everyone keeps a folder of screens they admire. Almost nobody writes down why each one works, which is the part that makes it useful later. When I save a layout, I add one line: what problem it solves and the one move that makes it. Months on, that line is what my gut is really reading. A folder of unlabelled images trains nothing.

1. 01Commit before you generate. I write down how the thing should feel, three adjectives and one sentence about the person on the other side, before I open any tool. Then I use the tool to explore inside that promise, not to find it.
2. 02Learn from outside software. Most of what I know about hierarchy came from magazine layout. Restraint came from product design. Pacing came from film editing. The design books everyone reads are shallow and shared. That is exactly why they cannot set you apart.
3. 03Ship, then stay. The feedback that trains taste shows up months later, in support tickets and retention curves. Leave before it lands and you learn nothing. Then you repeat the same mistakes, confidently, for ten years.
4. 04Defend one thing per project. Not everything can be special. Pick the one moment that carries the product's argument and make it undeniable. Let the rest be quietly correct.

The honest risk

There is a real danger here. Taste can turn into a shield. Some experienced designers use it to claim they are simply better, and to dodge tools that are making everyone faster. I have struck that pose myself, and it ages badly. The word becomes a way to win an argument without having to explain anything.

There is a second, sharper problem. A lot of what we call taste is really social signalling. The sociologist Pierre Bourdieu spent a whole book, Distinction, arguing that taste is often a badge that marks which group you belong to, not a judgement about quality. Design circles prove his point daily. Preferring a certain grid or a certain typeface can be less about the user and more about looking like the right kind of designer to other designers. When taste becomes a badge, it quietly stops serving the person on the other side of the screen.

Here is the test I use to tell real taste from dressed up preference. Ask what the choice is for. If the answer names the user, the context, and the result you expect, it is judgement. If the only reasons are that you like it or that it looks current, it is preference wearing a nicer coat. Paul Graham makes the same case in his essay on taste for makers: good design is not purely subjective, because you can point at principles it serves. The moment you cannot point at anything outside your own liking, be suspicious of yourself.

You can run this test on a Monday. In your next review, pick the three choices you feel most strongly about and force yourself to write one sentence each, naming the user and the outcome you expect. The ones where the sentence comes easily are judgement you can stand behind. The ones where you stall, or where the sentence is really about you, are the preferences hiding inside your taste. Cutting those is how you keep the library honest and stop it curdling into a set of habits you defend out of pride.

Dieter Rams said good design is as little design as possible. Back to purity. Back to simplicity. That is a call about what to leave out, made by someone who had shipped enough to fill a deep library and then trust it. No machine has ever made a cut like that on its own, and I do not think that changes soon.