The AI taste gap

AI can write code that compiles. It can't tell you whether the result feels right. Jony Ive: you might not always sense care, but you can always sense carelessness. AI-generated interfaces pass every functional test and fail that one. AI made code cheap. The artifact is what matters now. ## Why everything looks the same AI converges on the most common patterns in its training data. The most common patterns are mediocre. Every AI-generated landing page ships with the same Inter font, the same purple-to-blue gradient, the same three-column feature grid with `rounded-xl` cards. Not a bug. A statistical inevitability. The model picks the average. The exceptional is where taste lives. ## The "vibe coding" trap Generating a full page in one prompt feels productive. Speed without editing is how you ship something indistinguishable from every other AI-generated site. The problem isn't using AI. The problem is trusting the first output. Every generation is a first draft. First drafts need editing. The editing _is_ the craft. ## What the model can't judge AI has no opinion. It doesn't feel discomfort when a `border-radius` is wrong. It doesn't notice that the heading weight is too close to the body weight. It doesn't know that your users need reassurance on the pricing page, not features. Taste is the ability to feel that something is off before you can name the problem. AI doesn't have that sense. You do, or you will, once you've trained it.