The Library · Companionable Intelligence
The question behind the question
AidePersona asks what an assistant should do. This house holds the question underneath it: what living with one should be like.
Our sibling house AidePersona builds personal AI: assistants that adapt to a person across work and life. That is the maison's answer to what should the assistant do? SigmaClava holds the question underneath it: what should living with one be like?
These are different questions, and conflating them is how the industry keeps shipping presence without earning it.
Task competence is not presence
An assistant that schedules well has earned a place in your calendar. It has not earned a place in your kitchen, your arguments, your children's bedtime.
Presence, being ambiently there in family life, is a different grant of trust, and the interesting thing about trust at family scale is that it is not granted by the user. It is granted by the household, including its least-consulted members.
That last clause is where most designs quietly fail. A home contains people with different levels of consent and different levels of power: the person who installed the thing, the partner who was informed, the child who was not consulted, the visitor who never agreed to anything. A design that serves the first and ignores the fourth has not solved the problem; it has picked a side in it.
On the left, a territory made of slots that competence can book into. On the right, a household: continuous, tied together, and containing no cells to occupy. The dashed line is the one that task competence does not cross.
What we watch for
Our program studies presence as a designed thing with real parts.
Boundaries. Not settings pages, but lived boundaries. Who can the presence speak to, about what, when? Does the household know? Can a child tell where it ends? The working test is whether a member of the household can state what the system will not do, without reading anything, and be right.
Emotional regulation. Machines that respond to human feeling can amplify it or steady it. A presence in a home has a duty to know which it is doing. Our position is that when a person is dysregulated, the correct behaviour is almost never more input. That makes withdrawal a designed capability with the same weight as response, rather than a failure state.
Exit. A companionable presence must be removable without grief engineering. If leaving is designed to hurt, the design is extractive. The test is whether removing the system leaves any dependency the household did not knowingly accept.
Silence. What a presence does when it is not addressed is most of what it does. This is the largest surface in the design and the least specified in the industry.
Why family scale
We run these questions in our own home before we say anything about them. Family-scale deployment is slow and small, and that is the point: a household is the one environment where a design cannot hide behind aggregate metrics.1 Someone at the table will tell you the truth.
The program is active and internal. When something survives being lived with, it will appear here.
Notes
- 1Aggregates are not neutral here. A design that is slightly worse for everyone and much better for a majority looks like an improvement in the metric and like an imposition in the room. ↩
From the program
Companionable Intelligence
What does it take for an AI presence to earn a place in family life? Not to capture attention, but to hold trust, with boundaries a household actually controls.