AI-generated UX fails because it lacks anticipatory design. The interfaces feel wrong not because they are ugly, but because they do not predict what users need before they ask. This is the uncanny valley problem applied to software, and it is a gap that current models cannot close on their own.
The argument from Sidebar is specific: human designers understand intent, context, and the emotional arc of a user flow in ways that generative tools do not replicate. An AI can produce a screen. It cannot yet reason about the three screens before it and the two decisions that follow.
The full piece is worth reading for its framing of where AI tooling actually breaks down in practice, not in theory. The question it leaves open is whether this is a permanent ceiling or a training data problem waiting to be solved.
[READ ORIGINAL →]