I think I'd have a lot better shot with a semi-complicated Python codebase to be correct. Even with GUI stuff if using something like e.g. Qt box layouts.

Yes, CSS is a lot more powerful than those kinds of systems. Perhaps comparable to constraint satisfaction layouts that tend to be quite surprising.

But even so, there are so many settings and their combined behavior in CSS that it's, for me at least, totally infeasible to try to code much CSS "up-front" without constantly checking the result of the changes. And I don't mind it that much, but I'm quite certain that a more "predictable" equally powerful layouting system is entirely possible.

It could be an interesting challenge to show people some layouts and ask them to produce them using CSS without being able to iterate the results.