From what I understand, the way they work over at 37signals is that Jason Fried will do interaction design, graphic design, HTML, and CSS coding, which he’ll then hand to David or some other programmer, which will add code.
That is, indeed, different from my past experiences in many ways. It seems to me that typically a significant application involves significant up-front planning. You have a programmer work with the data first (the model), then work on the flow of the application (the controller), and finally you get some input from your designers (hopefully) when you build your view.
However, the folks at 37signals have long advocated working more or less the other way around. They design the interface, and how that is supposed to work, first, and then hand it off to programmers to take over and complete. Unconventional, but you can’t really argue with the results.