Shared family and team assistants

Multi-human, one-assistant setups:

• Role separation: who may approve spend, who may read HR mail, who may only ask for summaries.
• Shared lists and calendars with conflict handling (“we already bought milk”).
• Team rituals: standups, shop runs, kid schedules—encoded as lightweight workflows.

“Company assistant, family assistant, team tool” stories need permission design, not only features.

• Channels already meet groups where they chat.
• Approvals and audit trails matter more with multiple humans in the loop.
• Local-first options for households uncomfortable with cloud brain dumps.

1. Threat-model shared data: partners, kids, roommates, colleagues.
2. Create personas or roles with explicit capabilities—no implicit “everyone is admin.”
3. Default deny on purchases, medical, and legal sends.
4. Teach a shared vocabulary for commands—reduces mis-invocation.
5. Review monthly: who joined, who left, which keys still work.

• Consensus on what may never be automated in this group.
• Recovery if someone loses their phone—identity matters.

1. Start read-only shared briefings.
2. Add shopping and logistics with explicit budgets.
3. Add delegation to sub-spaces (kid vs parent channels) if needed.
4. Log sensitive requests for dispute resolution—policy-dependent.
5. Offboard cleanly when someone leaves the household or team.

• “Refuse this request because role X cannot approve spend.”
• “Summarize only what the whole group already agreed is shareable.”
• “Propose a calendar merge that respects privacy zones.”

• Kids or staff safety review for anything public-facing.
• Billing owner documented; no mystery subscriptions.
• Exit plan: export data and revoke tokens when the group splits.

• Over-sharing private messages into a family thread.
• Ambiguous identity—“tell the assistant” vs “tell the group.”
• Permission rot when someone’s role changes but keys do not.
• Social conflict mistaken for tech bugs.