Learning from this https://github.com/Eladlev/AutoPrompt and this https://github.com/razbakov/ai-secretary I am thinking about how different roles can write note to each other to enhance one another within an organization, instead of having all the roles to not talk to one another.

Only issue is that LLM agents are prone to "peer pressure" or "group think" so isolation of concerns + separated communication through docs is likely a good mitigation tactic, bureaucracy is likely functional.