Probably due to the extremely large amounts of React code in the training data set, not due to any inherent advantage of the framework.

LLM coding will lead to stagnation in languages and frameworks as developers start considering how competent their favorite agent is in each domain.

LLMs favour React and Tailwind because they’re popular, so they’ve seen a lot of them, not because of technical merit. This is the very manifestation of React winning by default and slowing innovation.

Given the quality of some of that combo I’ve seen vomited out, it’s technical dismerit.

I wonder if you’re better off with something critical mass enough that an LLM can correctly write it at all but not trained on mountains and mountains of slop.

This has been an interesting realization for me, seeing the LLM well poisoned by its poor understanding of hooks and effects has opened my eyes up to how much the average developer also doesn't understand these things.

I'm terrible at CSS and have been generating HTML/CSS prototypes with Claude. It really knows how to push out tailwind.