LLMs seem to work well with tailwind which I assume is because there was a lot of tailwind in the training data. As a human I never found tailwind to be particularly easy to work with anyway (mostly because I can't be bothered to learn another CSS framework) so I was constantly checking the docs when manually writing code. But I do like the rigidity it forces, which is probably a good thing for AI coding as well.
I don't really get this argument. 90% of Tailwind is shorthand for existing CSS properties. There's really not much to learn if you already know CSS.
I feel it's better for LLMs. For each class name, the LLM doesn't have to go out and consume CSS files and read all the cascading rules and determine what other classes elsewhere might have an effect. It has everything it needs to determine the styling and layout right in its immediate context