Rosenberg includes some interview answers that were deemed too technical:
You mean WinFS, the file-system revamp that was supposed to be a part of Vista and then got chopped?
No, a skunkworks Microsoft project — called Tesla. It was XP-based, an application, a layer on top, and instead of having folders, you had tags, and you could organize by that. If people wanted that, if there were demand for it, somebody wrote it well, we'd have it. We're very used to the Windows foldering system, which emulates a second order system. It has a third order capability that is unused, or not implemented well enough for people to actually use. There's also a cognitive step to be taken to get used to it. Not much of a cognitive step. We'd be doing this for photographs within Windows if Windows gave us good enough tools.
It's not a cognitive step, it's an economic step. To tag an item is to invest effort upfront for uncertain return later. The tags I assign today may not be the ones that connect in my mind to that kind of item in three months. As a result, it is not clear that tagging is worth it for any individual. Tagging is indexing in disguise, and good indexing is very hard. Tagging of images may be successful because there is no other good way of retrieving images. But the co-ocurrence statistics of file contents and of our interactions with them are there without us making any extra effort. Manual tagging may be a good solution for the unusually organized, but for the rest of us, search that exploits content and context better is where the sweet spot is. I never tag (let alone folder) most of my huge email archive, but I search it all the time. I might be persuaded by a good automated method for suggesting tags, though.
I like Weinberger's book a lot in his diagnosis of the situation, but I am not so convinced by his favored solutions.