In one of those odd coincidences, right after Dan Hill was discussing the ideas Tom Coates had about automatic id3 tagging, macintouch linked to AutoTagX. I promptly fired off a mail to both of them, but was still mildly surprised to see it turn into a post.
So, what's AutoTagX like when you actually use it? Well, it's OK. It's a little rough round the edges, especially for a version 2.5 (there are horrible redraw issues for the sheets, for example, and it really shouldn't be installing a CVS folder, or need a restart) but it generally seems to manage its major task, which is taking a minimal set of existing id3 tags and presenting choices from FreeDB to flesh them out.
The real problem is that the data in FreeDB, well, sucks. For example, this mp3 doesn't always get the correct artist, and that's entirely a FreeDB problem. Similarly, FreeDB has a number of matches for tracks from this Prodigy album, but they disagree on the total number of tracks, year of release and so on. Admittedly, they are better than what was there.
Sadly, CDDB now charge for developers who want their apps to access the database (which used to be freely available, way back in the day), and anyway, some of the same problems happen with CDDB.
Similarly, in the email I mentioned an app I remember using during January when I was using a Windows laptop. It turns out to be called MoodLogic, and it too wasn't quite the magic bullet. From what I remember it used its own custom DB, and the stick it used to get you to input data was that you could either pay or upload data to get credits. However, I don't remember the results as being that impressive, although I was probably throwing some really obscure stuff at it.
I suppose the conclusion of this is that, although apps exist that get some of the way to the automated library tagger, at the moment there's still a lot of work to do on the databases that people use.