Useful content + Ontology description language + triplet tags + visual tags + UI code + personal search agent. Am I getting close? Encapsulate objects on data entry. Modular xhtml in presentation templates. Move along relationship arcs or standard top/left menu bar interface. Winning is the most efficient tagging. What's the best way to tie in executable code? Remote portlets? Exactly how bad can they muck up the equation?
The agent is pure software.
It seems to be the way things are going. There has been a lot of movement in the semantic web lately, and it looks promising. First you define your ontologies. Then you tag your content. Then you build a personal search agent to your preferences. Will it work? Some very smart people seem to think so. It's an awful lot of tagging, and the interplay between ontological relationships and the constantly changing web content is hard to fathom. For things like people directories it's straightforward enough (Jill is a member of HR, John is a member of Jill's team, this is Jill's phone number, etc) but what about scientific research? Is it just a brute force memex?
And if it does work, why hasn't it happened yet? The first barrier to entry is the engineering of ontological processing engines. Ontologies exist out there, and are easy enough to define, but to make them machine-processable, is a challenge. Then comes the monumental task of tagging all that content. Sees daunting, but the task of building a web site for every business in the country once seemed impossibly huge, also. If it's useful to people, they will do it. And may the best ontological tagging graph traversal engine win.