My prediction is that the kernel of the idea that will make GPG usable is to dispense with the idea of a single keypair, and instead build features that generate ephemeral keypairs on the fly. Make the system workable for users even if they don't understand what a keypair is. Some of what makes OTR effective can be implemented using PGP as the underlying cryptosystem.

When one suggests replacing OTR, one tends to get an earful about the importance of forward secrecy. I think forward secrecy is very important for systems in which there are extremely high-value keys that are "stationary targets". I think forward secrecy is less valuable in desktop applications, where the attacks that would cough up a persistent key would tend to be devastating to the whole cryptosystem anyways.

It's also worth saying that PGP isn't a particularly great cryptosystem. "Modern" PGP predates a lot of important stuff in crypto. But it's a very well studied cryptosystem.

There are strong cryptographers who are working on much, much better systems than PGP. The problem is that those systems will compete with amateur systems and the winner won't be chosen by security. At least with PGP, we know what we're getting.