As I said to someone the other day after dealing with yet another failure by Dropbox to do the safe thing or even the reasonably-recoverable thing with our data: "One reason I'm not worth a billion dollars is I would never have let Dropbox ship like this!"

Doing sync on top of an existing filesystem, making it work with existing apps, maintaining solid transaction semantics, and having a simple, understandable user model -- you just can't do all four. The Dropbox experiment has now shown that "correctness" doesn't matter, commercially speaking.