To give a concrete example, this would be useful for any Ethereum app that doesn't want to store data on a central server (which most cannot do whether for legal, security, or ideological reasons). The idea to me is that the internet wasn't built very well to run decentralized apps [which is definitely the case if you've ever tried building one without having to rely on central servers for caching, storing accounts, comments, etc.]. It's, imo, a nice complement to blockchain tech like Ethereum and Bitcoin. Long term once it's out and running I see dapps like Augur [a decentralized prediction market which I work on - http://augur.net] using it so users can securely store their private keys, report data, market data, trade history, etc. and easily go across/between devices as opposed to just using localstorage [which is a pain to migrate using] or fetching it from ethereum every time [which is very time consuming and has lots of overhead].
If we're going to seriously move in this direction of decentralization, at scale we need something like urbit. No one else is really tackling the same set of problems.
Came across this quote on it by Alan Kay: "They have verve, and that's generally a good thing. In this case there are a lot of details that need to be grokked to make any reasonable comment. The use of combinators (a kind of dual of lambda calculus) harks back to an excellent thesis by Denis Seror at the University of Utah in the 70s that produced a safe, highly scalable and parallel implementation. I haven't looked at it more deeply (and probably should)." [https://news.ycombinator.com/item?id=11810177] - very cool!