You don't refer to bananas as "the curved sweet yellow food that grows on trees", you simply refer to a banana as a banana.
Sure, for one off items it might not make sense to name it but when you create design patterns then it's incredibly useful to be able to name things.
Yes it may be hard and require you to think, but just because that is so does not make it bad necessarily.
You have (some variety of) cavendish bananas in mind. There are red, pink and blue bananas. They're not all sweet. The plants they grow in are not always "trees" but rather large fern like plants.
And then that day arrives when you are required to style some variant of 'banana' you are forced to decide: do I break the world by messing with banana or do I make new-banana?
In CSS land however you wouldn't break the world for a different banana, you'd simply add another class for the variant and style appropriately.
.banana { // generic banana styles }
.banana.musa-velutina { color: pink; }
Of course not all decisions are going to be as straightforward, but I do think that there's a lot of value in keeping a website's items consistent.
The work upfront to name things and build out concepts will pay its dividends as you extend a site or build out new ones.
If you plan on using bananas multiple times throughout your app, aren't you going to be pulling that out into a component anyway? So define that it's curved and sweet and yellow in that one place, and then insert an instance of it wherever necessary.
Of course, there’s no right or wrong answer here, and I suspect it very much depends on what type of web systems you are building. I feel the entire debate is basically a special case of the question “do you make websites or web apps”.
If, instead, you litter your code with my-brilliant-semantic-class you create a different kind of problem. Requirements change and my-brilliant-semantic-class won't be sufficient. Now the choice is; alter my-brilliant-semantic-class and suffer all of the unintended side effects or abandon reuse and make another-brilliant-semantic-class.
The likely choice is the latter and so applications become masses of ad-hoc, partially finished, inherently flawed styling abstractions scattered hither and yon among directories full of redundant styling artifacts, half of it long dead. How is that a benefit to some future re-designer?
> next time a designer decides to give your webapp a makeover?
I figure the odds of one approach vs the other being the greatest benefit to some future re-designer are about even, and my guess is as good as yours.