Interesting elements: desktop integration (peristent storage, UI touchpoints, etc.), full HTML(via WebKit) support, integrated PDF display, and an application packaging model for easy install and update (distribution).
The two most important pieces (IMHO) are the persistent storage model and application packaging model - these start to bring the benefits of "content" to executable applications... that its "Flash" underneath is interesting but not necessarily the important part. These technologies do provide a differentiated but overlapping feature set that I don't mean to downplay, but HTML, Flash, PDF, etc. are, I think, going to end up being more about onboarding developers.
Hard to appreciate how important distribution is as a part of application lifecycle management.
The natural comparison points are Silverlight and Prism, and as with those two, distribution will be less the issue than finding the compelling applications will be (although you should note that Silverlight, in particular, is more an answer to Flash than AIR).
Adobe is significantly advantaged in this regard because (a) they're cross-platform (by that I mean crossing OS, browser and Web/Desktop boundaries), and (b) they're committed to the platform - the biggest knock with Microsoft and the Mozilla Foundation is that its hard to tell how serious they really are. Remember when all of Vista was going to *require* .NET? Or remember ChromeEffects? Or Blackbird?
That said, I think one chink in the AIR platform (IMHO) is that it provides no way for third parties to natively extend the platform - Adobe still thinks about applications as B2C propositions. A second knock is that they still haven't gotten the "just in time" element of application and platform install quite right (I've written on this before) - it will limit (again IMHO) where users prefer websites to applications.
Still - there's a lot to like.
More shortly - Kevin Lynch is talking now...