What criteria or triggers should determine the creation of a sub-agent within the system?

[abhay97ps]

I run into this question when thinking for system to efficiently balance the need for structured, periodic updates with a vector db to align and sync with other agents. Does the requirement for higher order agents is to maintain logs? Essentially when context is getting large to handle new agents are spun up.

Tasks agents at leaves nodes generate most logs and will require quicker iterations. They rely on last updated information version. Where higher other agents track progress, and manage requirements and help resolve issues. These agents require a more longer term temporary memory and multiple iteration to consolidate information.