Problem
Multi-agent systems need shared planning, but shared execution context can become chaotic. If everything is shared, no one knows which agent owns which state.
Short workflow
- Keep each execution agent in its own identity.
- Use shared workspaces for planning artifacts that truly need collaboration.
- Reference identity-specific work from the shared workspace instead of merging inboxes.
- Review progress through shared notes, calendar, and kanban only where coordination matters.
What belongs in shared planning
- Cross-agent task boards
- Shared milestone calendars
- Coordination notes and decision logs
- Work that several operators or agents need to see
What should stay identity-scoped
- Inbox state
- Agent-specific credentials or tokens
- Sensitive workflow notes that do not need collaboration
Decision rule
Share planning by default only when multiple agents need the same context at the same time. Keep execution isolated otherwise.