Getting a prototype up for LangChain to generate UI variations is not super hard; if you could just scrape and index a bunch of sites, I think you're off to the races.
With that said, I think it can't replace a human in the mix (yet)
I wonder is there also a factor that - since there are so many low/no-code and WYSIWYG UI tools on the market already - the potential time savings are lower compared to back-end tasks that are being targeted by AI tools - ie providing language prompts to get a tool to output UIs and then needing to tweak them to implement your own branding/messaging/design preferences might not be that much faster than building the same UI from scratch with the kinds of solutions already on the market.
Very interesting to see how this space could play out over the coming years anyway.