It would ("instantaneously") generate an automatic set of tags based on page content and metadata. This could very rapidly edited in the form of a simple, comma-separated list of text.
Bookmark search consisted of typing in a simple, comma-separated list (IIRC). The current contents were continuously parsed and partial matching performed to produce a ranked list of matching entries. With thousands of bookmarks, the desired entry could usually be identified and followed within a second or three.
(There were also date fields and some other fields, that could help in clustering and could be used in ranking)
The combination of a tagging categorization, and the very efficient interface, made it a joy to use and extremely useful.
This thread's product appears, from the video, to be moving the interface portion of web-based tagging in what for me would be a "right" (correct) direction. Make the tagging rapid, and the interface painless, and perhaps tagging will really take off as an end-user interface.
Note: I followed vijaydev's link, as the ur-parent's link is currently unresponsive.
I suppose using this for tagging stuff has two positive side effects:
1) You can now use longer tags, with whitespace, capitalization, etc.
2) Duplication of tags is much smaller.
Kudos to the author.