In the aftermath of RailsConf, one of the most interesting ideas came from Evan Weaver - later echoed by Dave Thomas.
The idea is that SQL itself is a kludge. I read somewhere that DHH said at RailsConf in 2006 that he uses databases as hashes. Evan Weaver went one step further and said that there are very few Web apps where you don't need a hash instead of a database. His suggestion was that SQL itself is a design mistake.
People use relational databases to provide persistence in Web apps, but they don't do it because databases exist for that purpose. It's a little like using a necktie as a belt when you're late for lunch. The Web came along, and everybody suddenly needed persistent data yesterday. SQL databases existed, so everybody tied them around their waists and ran off to lunch.
Avi Bryant's keynote presented just one alternative to the necktie belt - the object database, handled by a virtual machine. This is a powerful, successful option used in performance-intensive financial applications.
The future of Web apps will probably bring us more options. There will be a lot of apps using SQL for a long time to come, just because "that's how everybody's always done it", but the case for GemStone-style Web app VMs is a strong one. One of the major unifying points among Seaside, Rails, and the entire armada of new Web frameworks released in the past two or three years is that they all minimize the amount of raw SQL you write. Already, the fact that you can write raw SQL is becoming a differentiator for Web developers.