Great overview! I’m curious how you’d approach managing state and long-term memory in a multi-agent system when agents need to share context without introducing bottlenecks—are you leaning toward a centralized blackboard pattern or something more distributed like event sourcing?