Have you ever seen the movie Demolition Man?
It's set in a future where every restaurant is Taco Bell.
Taco Bell was the only restaurant to survive the Franchise Wars, so now all the restaurants are Taco Bell.
When I think about the current state of LLMs and code generation I think about Demolition Man and Taco Bell.
If you've worked with LLMs like Claude and ChatGPT or tools like Cursor and Bolt you will have probably experienced the huge improvements they have experienced over the past year, especially around generation of UI/UX.
It's quite impressive, you can use an LLM to put together specifications for new screens (or for an entire product) and then feed those back in and magically you will see a React UI appear before your eyes.
LLM's can drive tools to design UI. I don't think they are even close. But they are incredibly effective at pulling together React components. It's the way that they are most effectively able to communicate design.
Simply put, LLMs most effectively communicate design by programming React.
With what we've seen in the past year, I'd be very surprised if going into 2026 we didn't started to lose diversity in front-end languages because LLMs are so much better at conveying design in React. Everything is going to start looking like shadcn/ui because that's what the LLMs do best.
Don't take my word for it. Make a project in Claude, or build something with Cursor or Bolt. Start a new project with Replit. Check out Vercel's v0.
As we wrap this all with agents humans are less and less involved in the decisions around the creation of things. I think we will just naturally see an exponential increase in React. Specifically React components that leverage TailwindCSS.
If nothing else this will become the defect way to prototype things. It has to. It's so quick. Which means every prototype is going to be built of the same underlying pieces. The same components. Even if it is then translated into another language before being deployed, its heritage is React.
In the future every UI is React.
One final thought on this. I can't help but wonder if we've set ourselves back years when everyone decided that Tailwind was the best way to build things. Previous to this frameworks like Bootstrap and Foundation conveyed meaning, but Tailwind is really CSS as a class in HTML. Given how well LLMs interact with HTML I suspect that they would have been more successful in "understanding" intent if we labelled things as buttons.
Member discussion