I might reconsider it again once I am more comfortable with what frontend has become, but that is still quite a bit away.
A big part of Tailwind's value is that it provides a curated, well-documented subset of the most important CSS attributes. The utility classes are consistently and concisely named (unlike many native CSS properties, which grew organically).
You're totally right that it is yet another abstraction, and that can be tiring. Personally I found it to be well-designed and transparent enough to make it worthwhile.
I'm guessing most of the naysayers here work in the web applications world with component-based architectures, where the HTML/CSS is a small portion of the total time they spend coding, so they can't understand why someone would sacrifice readability, maintainability, and organization for speed on something that isn't a bottleneck for them. The thing is, it absolutely might be a bottleneck for agencies and other companies that need to build lots of custom sites.
And the argument is that this comes at the expense of maintainability.
It seems like if your project is not suited to a "website bashing" or prototyping approach long term, and you need semantic classes to allow for easy theming type use cases, it could be more trouble than it's worth?