The difference between a product that works and a product people choose

Making a convincing first version is cheaper than it has ever been. A model will give you screens, a flow, a working prototype, in an afternoon. The thing you would have spent a month proving is now an afternoon of prompting.

That changes what people notice.

When several products promise the same thing, the one people pick is usually the one that is easier to understand. Not the one with more features. Not the one that shipped first. The one where you did not have to stop and work anything out.

The clearest example I know is the one in your pocket.

Bluetooth pairing was solved for fifteen years before AirPods. It worked. Settings, scan, find the device in a list of devices, pair, hope. Everyone had quietly agreed to accept that. Then AirPods put a card on the screen the moment you opened the case. The engineering underneath was difficult. For the person opening the case it was simple: the next step was already in front of them. Less to work out.

That is craft. Noticing where someone has to stop and think, and then removing that moment.

Craft is deciding what to ask, what to leave out, and what happens when someone gets stuck. It includes how a thing looks. It also includes whether the next step makes sense. Get the first part wrong and no amount of polish on top saves it. You can feel this in any product that looks beautiful and still annoys you by the third screen.

A polished screen copies in a day. Learning what to change next takes a year of watching people.

That is the real difference, and it is worth being precise about, because "have better taste" is not advice anybody can act on. A competitor can copy your screen by Friday. What takes them longer is the two years of being wrong that taught you which screen was right, and the habit of going back, watching people use it, and changing it again.

Here's the honest pushback, and it's a fair one.

Taste by itself isn't a moat. Real moats are things a competitor can't just copy: users, data, distribution. Taste is a skill. You can hire someone with taste. If your whole argument is "we just have better judgment," that's weaker than "we have ten years of user data."

The sharpest version of this I have read is Dan Maccarone's. His case is that the destination is the problem, that people are loyal to the intelligence and not to the thing you built on top of it, and that a long list of well made AI products churned anyway. That last part is simply true, and it should bother anyone making my argument.

The honest answer is narrower than "craft wins". A better experience cannot rescue a product that fails at its main job. Nobody stays for a beautiful thing that does not work. But capability on its own does not settle it either. People still have to understand what the thing can do, get something useful out of it on the first try, and recover when it gets something wrong. Those are design problems, and they do not go away when the model improves.

So here is the part you can actually use. The next time a prototype works, watch one person use it without explaining anything first.

Where do they pause? What do they read twice? What do they try, undo, and try again? That is not a bug list. That is where the next round of design starts. A model can tell you what might confuse someone. Watching them tells you what actually does.