Of course, Ingo has a point by demanding "a light-weight source language" and "getting out of the way". (Hence the popularity of markdown.)
I fear we would have to go into a lot of detail about my workflow - and in a way, that's what it comes down to: use cases, habits, requirements are all different. Going from mdoc to DocBook wouldn't work in my case because 70 macros is just too limited. But the way I write is similar to what @alex describes: I start jotting down something in AsciiDoc, have it converted to DocBook, and add elements as my ideas evolve.
mdoc, DocBook, etc. all start by separating semantics from presentation, and that's a good thing. But often the visual presentation helps to make a dense text jungle penetrable. And that's where CSS comes in handy. Not as something to make text appealing ("spray paint and crayons"), but for structuring multilayered text.
Does that make sense?