One of the big challenges for all of us developing Web 2.0 is finding out what the users actually want. That's part of why libraries and publishers are taking various lab, web development group, developer network and contest approaches. Another big challenge is trying to build a community around your tools.
You can read more in his posting Engaging a New Generation of Authors, Editors, and Reviewers - Presentation for Boston Editors' Conference.
One of the big challenges in this area is getting people into the new network in order to collaborate - this is the issue that most of the "scientist social networks" have - there's no one there. Even using Google Docs to collaborate tends not to scale very well, as you end up having to invite and reinvite groups of people to each new document. FriendFeed as a collaboration space I think has a lot of advantages, in that it brings an immediate, individual benefit (mindcasting/lifecasting unification of web activity) as well as a critical mass of people, as well as lightweight group and discussion tools.
The key point is it can no longer be either/or. It's not my site OR your site, it's all of our sites, interconnected. It's also/and. I have to be able to go seamlessly from LibX to Zotero to Mendeley to Connotea to EndNote to 2Collab to Papers to FriendFeed and back again. If Zotero has a great tool, I should be able to call it using an API from LibX. Data can't be trapped at point of creation, it has to be able to flow around the web to be useful.