Log 008: Household Gods: Permission Layers in Domestic AI¶
Domestic-scale orchestration where the household is a contested multi-agent system.
- Mode: Thinking Space
- Status: Working Note
- Date: April 2026
- Scope: domestic orchestration and permission layers
- Depends on: Chord vs Arpeggio, Epistemic Firewalls, Human Vital Systems
Problem¶
Most assistant architectures treat "the household" as one user. This is wrong on contact. Homes are plural polities: unequal power, uneven memory, conflicting consent.
Proposal¶
Model home assistants as permission spirits rather than omniscient servants: - memory steward (what is remembered, what expires), - boundary steward (who may ask what, when), - escalation steward (what leaves the house).
The key design principle is non-unitary consent. "Family preference" is often a coercive aggregate.
Failure cases to design for¶
- Grief lock-in: dead relatives become permanently active through never-expiring memory loops.
- Domestic capture: dominant member weaponizes assistant access policies.
- Care/control inversion: safety checks become surveillance theater.
Architectural consequences¶
- Require per-member veto lanes and temporal consent (revocable later).
- Default to partial forgetting, not infinite retention.
- Use local-first logs with explicit export rituals, not silent cloud drift.
Open design question¶
What is the minimum review bandwidth for a household to remain governable without externalizing every intimacy to platform arbitration?