I also prefer the 'bounding box' paradigm. I'm so used to it now that I can't stomach the idea of modeling in anything that doesn't automatically give me my little scaling box.
As for the short-cut driven versus mouse-driven, I believe what was meant is that a program must first build a model in the user's mind and this is (imo) more easily done visually. Once that user becomes a "power user" or "experienced", however, it is much easier to tack shortcuts onto the existing mind model and thereby speed up workflow. All developers should assume an iterative path from beginner to mastery unless they develop things that already have established UI behavior. (Such as a word processor)