The problem area attention.xml fits in to (if I understand it properly) is improving blog (or really any feed content) reading efficiency by helping to prioritize entries and reduce duplicates.
Just under a year ago, I suggested Bayesian classification for blog reading prioritization. My idea then (resembling an idea I had ten years earlier for reading USENET) was that your reader would predict, on the basis or what you read (or marked as interesting) what other posts you are likely to be interested in and prioritize accordingly, using Bayesian classification much like spam filters. My idea was not that entries would be filtered out nor that new entries from unsubscribed feeds would suggested to you. The idea was just to help with prioritization.
It seems like there could be a lot of synergy between that idea and attention.xml. I need to think about it some more - watch this space!
I certainly think there are still massive opportunities for innovation in blog reading technologies.
And where might Leonardo fit in? Given that I see Leonardo as the "hub" of my online presence, a lot. The key will be how to better integrate my feed reader with Leonardo to enable support for things like attention.xml