This is similar to something we've been doing for a while. Instead of individual agents we are creating many iterations and sub-iterations of spawned agents that are largely autonomous. A lot of the human-centric paradigms just don't really apply to LLMs/AI but people are used to approaching them that way.