Case Study 01 · 2026
Living Home
A family's home, redesigned as a shared decision system.
The fridge already stores the family. The calendar already knows the week. The car already knows the route. The problem is that none of those surfaces know how to decide together without crossing a line.
Living Home uses Context Grammar to turn household context into governed action: what the system may remember, what it may infer, what it may do quietly, and when it must ask.
The premise
Samsung Family Hub already knows allergies, dietary restrictions, and family profiles. A home AI can already generate recipes, shopping lists, reminders, and route suggestions.
The hard product question is not intelligence. It is shared judgment.
Five people live in the same home, but they do not share one permission model. A parent may see budget context. A child may own music preferences. Health data may shape dinner without being exposed to everyone. A $2 yogurt can be auto-picked. A $150 bottle should wait.
What the case proves
- A family AI is not one assistant. It is a set of agents coordinated by one household brain.
- Memory is not a log. It is permissioned context that changes by person, domain, risk, and time.
- Autonomy is not a toggle. It is governed by stakes: cheap, reversible actions can move; sensitive actions slow down.
- Multi-surface UI needs one semantic state rendered differently across fridge, phone, car, speaker, and family archive.
System anatomy
Living Home separates the system into three layers.
- Context Brain: preserves family memory, relationship rules, permissions, and long-running preferences.
- Coordinator: reads intent, signals, dials, and trust constraints before deciding what should happen.
- Surface agents: render the same decision state into the right interface for the moment.
This is why the home can behave intelligently without pretending every surface is the same product.
Proof moments
Morning kitchen
The fridge resolves three different lunch decisions from one inventory state. Mai gets the fastest default. Sota gets the familiar safe option. Leo's dairy constraint changes the substitution path without turning breakfast into a medical disclosure.
Dynamic friction
The same grocery system treats different stakes differently. Low-cost, reversible items can auto-pick. Higher-cost or socially sensitive choices pause for approval. Trust is earned by reducing unnecessary decisions, not by asking about everything.
Permissioned agents
Finance Agent cannot read grades. Education Agent cannot read the budget. Aoi can update their profile once and every agent's access boundary rewrites from the same source of truth.
Care architecture
When the school nurse calls, the car surface collapses to one focal action. The system does not show the whole brain. It routes the minimum context needed for care.
Design decision
The UI is designed around traces, not magic. Every decision leaves enough evidence for the family to understand why the system acted: which signal mattered, which rule constrained the action, and what changed across surfaces.
Tradeoff
Living Home adds a serious metadata burden. Preferences, permissions, reversibility, sensitivity, and household roles must be modeled before the interface can feel effortless.
That cost is the point. Without it, "smart home AI" becomes a collection of confident shortcuts. With it, the home becomes a governed decision environment.
Outcome
The flagship argument is simple: agentic AI becomes useful at home only when it can read family context without owning the family.