An interesting proposal from Peter St. Andre on where Jabber should be going, he's suggesting that the core protocols are basically sane (I've not dug into them in enough depth to find significant holes in that argument), and that what's holding Jabber back is the lack of a standard client and server.
His idea is because so many of the current servers and clients appear to be small-ish half finished projects, and concentrating the efforts from these dozens of projects into one master project may be far more fruitful. I'm thinking tower of Babel, but you never know, with a little co-ordination and ego-soothing this may be possible.
I think some of the current clients are actually maturing nicely, Psi and Exodus are certainly very usable, so I'm not sure about the demand for a standard client. However, perhaps something simple and moderately lightweight would be ideal as a introductory client. I myself used Jabber Inc's old client for a long time because it was simple and worked.
One thing that Peter suggested that does sound really useful is for the standard server to be aimed at a small to medium size organisations, and I also think writing this in Python is a good choice. Yes, I'm a Python fan, but I'm also a C++ developer and I'm thinking of the multi-platform issues, which Python has far fewer than many other languages.
David Thomson's also got a few thoughts about the proposals which follow on quite nicely from his earlier thoughts about Gibber (Google and Jabber). Peter follows up too, and he also seems to be veering away from a standard client a little.
One further thought on the standard client/server, should there be a reference implementation of the Jabber standards? I.e. from a users' point of view they wouldn't necessarily be the nicest or best to use, but for a Jabber developer the protocol handling would be a textbook example of how to do things. I suspect a well designed and very usable fully standards compliant client/server would be more far better.
Stop Press: Having just talked dwlt through connecting to a few transports with Psi it seems to me that the requirement for a simple easy-to-use working client is paramount.