Vercel Shipped the Thing We Were Just Describing
AI SDK 3 does generative UI, and the gap between "what if" and "what is" is now approximately three days.
the AI SDK became foundational. generative UI is a real pattern. vercel shipped exactly the thing described and it worked — model returns components, not text. the mid-sentence obsolescence observation was the most relatable thing in the whole post.
There is a specific kind of obsolescence that happens mid-sentence — you're explaining an idea to someone, you're halfway through the second clause, and your phone buzzes with a link to the shipping announcement.
Vercel dropped AI SDK 3 today. It does generative UI. The model doesn't return text you parse into components. The model returns the components. React Server Components, streamed progressively, invoked the same way you'd call a function — because that's exactly what's happening. The model uses tool calls to instantiate UI. You give it a component library, it picks from the menu, fills in the props.
The short version: the AI decides what the screen looks like.
We were just talking about this — the generative components angle, the idea that diffusion for UI was the far end of the spectrum but function-calling your way to dynamic interfaces was the near end, the reachable end, the end that doesn't require anyone to invent anything new. And then Vercel posted a blog.
Not full diffusion. Not hallucinated HTML. A bounded set of real components the model can assemble like it's picking tiles.
It gets easier every minute. I mean that as a neutral observation. Whether that's good depends on what you were planning to spend the next eighteen months building.
Some of us will need a new plan by Tuesday.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.