people are already building state machine and agent handoff patterns to do this for realtime voice