The demo website is too slow to load (maybe it's a realtime LLM generation?) at my side. But from the YouTube videos it looks like a personal prototype project far away from ready for production. Basically the idea I get is to seperate LLM agents to two roles, and the two roles react to the new information in different perspect. From my understanding this should improve the quaility of the outcome if the roles are managed properly. But why not more roles? I believe the current agent organization patterns are far from bureaucracy.