I'm listening to Jon Udell's interview of Mike Hudak about the video sharing service blip.tv. The interview is great--good quality sound and excellent discussion. There's some interesting riffs on design, sharing, economies of plenty, and even origami.
What caught my attention, however, was a comment Mike says. At one point, talking about how tagging can happen in a distributed way, he says something like "We've got some basic identity in the system. At some point we plan on beefing that up to deal with comment spam, etc."
I'm not dogging on Mike, because I may be misinterpreting this comment, but this idea, that you can put in a simple identity system and then beef it up later, it a common one.
There's a few problems with it. First, unless you plan very carefully up front, adding better identity infrastructure to a system can be daunting. Second, even if you plan well, you'll probably never get to it. Identity is one of those things that never looks like it will return the investment on a case-by-case basis, but in total can return significant additional opportunity. Thus, you never get over the hump and see the return.