Somehow I keep coming back to "Notes on the synthesis of design", in particular the need to partition problems to control the complexity.