Ah, ok. Visibility into subagents makes perfect sense. From the "reddit-style" description I was imagining something where the user was conversing non-linearly with the agent at different points in some big tree.