Earlier quoted context omitted.
You're not wrong, and I mostly agree with you. I die inside when I see the div soup that a lot of sites have become. However, I think there is value in being able to have the important parts of CSS merged into the HTML a bit. Where that line is, is certainly up for debate (and I don't have the answer), but I've found a lot of my tailwind sites are more readable to me than my pre-tailwind sites, often because I don't…
The "open a different file" reasoning piece is a common pro-Tailwind statement and I do see the upsides. I think that upside became more prevalent in the reusable components era, whereas previously CSS was targeting an entire HTML file (and thus the reasoning was more like SQL query than "this one element's styling"). With LLMs I think this upside is much smaller now though.
With LLMs Tailwind wins. Because it's a very restricted set of classes. With regular "separation of concerns" CSS, LLMs will happily just pile on more and more and more CSS because they can't really analyze the code that's already there, and will miss and re-create huge chunks of CSS. Or write increasingly hyper-specific CSS to fix reported issues.
Anecdotally: in a side project I now have 10k lines of "pure" CSS generated by LLMs on top of Tailwind. The web part of the app is ~20k lines (not all of them are rendering anything on screen). No idea how to fix it :)