Generative UI Doesn't Make Sense

Hamel Husman makes the case that building UI on the fly from LLM output — what the industry calls "generative UI" — is mostly a solution looking for a problem. The idea sounds seductive: an LLM writes the components, the layout, the styles, all in one pass. In practice, it's a fragile house of cards. One hallucination and your form breaks. One context drift and your buttons shift to the wrong side of the screen.

The real insight is that UI has structure. Good UI is hand-crafted components — buttons, cards, modals — that work together reliably. When you generate it all, you lose that structure. You get something that looks right but breaks under real use. The fix is to use LLMs to select from known components rather than inventing new ones, or to constrain the generation so it can't break what works.

The broader point: not everything benefits from being generated. Some things are better built once and reused. Generative UI isn't wrong — it's just overhyped. For most apps, the right move is a small LLM call that picks the right component, not a full generation pass.

Why this matters for us: la gente building real tools — the ones that survive contact with actual users — should resist the urge to generate everything. Build the components once. Let the LLM do the picking, not the inventing. The tools that work are the ones that don't break when you're trying to use them at 2pm on a Tuesday.