Scaling from one user to a team is where directives and procedures have to be separated cleanly. We learned this at empla.io, conflating language and style rules with action rules produced contradictory systems that drifted fast. The code-oriented agent in our fleet stays reliable because its domain is tight, the executive assistant starts drifting the moment scope widens past two or three focused communications a day. Did you end up with shared system prompts per team or per role, and how did you version them across users?