I enjoyed his talk. It was deliberately provocative, and perhaps not even Ted believed all of it. But definitely worth hearing and thinking about.
Ted's main point was that most modern computer systems (both the OS and the applications) today are essentially the same as was defined in the 1970s on UNIX. That time (the birth of UNIX) represented a moment (perhaps the only one we'll have) to break free of certain concepts. He discussed paper - and how it is a constraining medium for information - look at all the funny notation that copy editors and others have for denoting meta-information. Look at the problems people had in versioning paper documents. None of this is easy with paper and we knew these problems. So in 1970 we had an opportunity to re-invent the document. But we didn't. We still essentially imitate the paper format. [And lots of other weird restrictions - why is a document a single file?] As Ted put it: WYSIWYG is just propoganda for paper simulation. (Or better: doing this is the equivalent of taking a 747, ripping off the wings, crippling the engine and using it as a bus). As HTML contains embedded markup (and only one-way links), Ted classifies it as an "unmitigated disaster"!
He demonstrated the Xanadu Space application, which tries to offer a fresh look at documents. It looked quite nice, although not terribly novel (I've see similar apps for photo management for example).
He also described design in computer applications by considering the representation of time. We tend to show it using a circle for the hours and a table for the calendar. He pointed out that we should really unify these views and represent time as a spiral. He also made the following point, which is really quite beautiful:
The aim of any user interface is mental clarity
He wrapped up his talk by discussing very briefly his generalization of data structures: hyperthogonal cell constructions. This was quite nice, although I didn't agree with all of his criticisms of relational databases. First of all, Codd's vision was relations, not tables. Second, whilst we accept that tables are an artifical construction they allow us to build expressive query languages and execute them (very) efficiently. It's a trade off. Ted didn't mention any programmatic querying of his data structures - but I'd be interested to see if there was anything plausible here.
Anyhow, a fun talk.
For the interested: An entry on Ted on wikipedia is here.