Do you mean inline in CSS / style tags, avoiding visual HTML classes (in which case we're in agreement - there's one way to edit how something looks, though style tags have other issues) or combining either of those with visual HTML classes like this library uses?

Okay, I think we're in agreement and are just crossing wires a but. The main reason i'm loosely okay with Semantic UI is that I just see it as using HTML fragments as building blocks rather than using JavaScript components. I wouldn't advocate it for anything elaborate, but I think it can work well within a certain problem space.