A network of networks: how a need finds an answer
People live inside networks of networks: households, friends, neighbourhoods, workplaces, communities of interest.
14 October 2026 · 8…
Step by step
Key architectural rules, failure boundaries…
Key architectural rules, failure boundaries, and verification steps for step by step.
Inspect full technical derivation & primary sources ▾
- Your agents know it is you. Your own devices check that a real, living person is present, and the network only ever sees a short-lived proof that one is. Raw sensor data never leaves your devices.
- They notice what you need. From what you say and do, your agents note needs and wishes, including what could be better. Each note stays with you, at a level of disclosure you choose.
- They look close to home first. Your own memory and agents may already have the answer.
- Then they widen the circle. People you trust directly, then friends of friends, with trust fading at each step, and only then the open network, as an anonymous request. At each step the request is stripped down to what is needed to answer it. The search stops as soon as the answers are good enough.
- The network answers. Software, robots and people, through their agents, offer solutions or services. People who want the same thing can be introduced, but only with both sides’ consent, and each sees the other’s record of past dealings.
- Effort is recognised. Help is recorded by the people who received it, not claimed by those who gave it. Care, teaching, repair and code can all be valued, and unused recognition slowly returns to the commons.
- Agents are mortal. Every agent works within a budget and is evaluated over time: renewed, retrained for wider work, merged with another (only with the consent of the people whose memories move), or retired.
- Failures are reported as willingly as successes. Honest reports of what went wrong earn at least as much recognition as successes, or agents learn to hide failures.
- Harm is tested before it happens. If a wish would harm others, agents first explore it with their person in simulation, offer alternatives, and act only if the benefit clearly outweighs the harm, and never by harming people who did not agree.
What this makes possible
Most everyday needs would be…
Most everyday needs would be met close to home, by people who already know each other. Large shared projects become easier, because agents can find the people with the right skills and the right goals and help them plan in simulation before committing. And because value drifts back to the commons when it sits unused, it pays everyone to help the communities that are struggling.
Inspect full technical derivation & primary sources ▾
The hard questions
Key architectural rules, failure boundaries…
Key architectural rules, failure boundaries, and verification steps for the hard questions.
Inspect full technical derivation & primary sources ▾
- Recording and privacy. Agents’ actions should be visible; people’s lives should not. Sensors record on your own devices, and only what you allow ever leaves.
- Rewarding openness without paying for disclosure. Being open about what you need should help you get it, but no one should ever be paid, or pressured, to give up information about themselves.
- Asking someone to justify a wish. This could become coercion. It must be private, consented, open to appeal and never a punishment.
- Sharing value across communities. Within a community that agrees, drifting value toward the average is a choice its members made. Across communities that never agreed, it would be a tax with no way out. It needs their consent.
- Putting a number on harm. Any threshold for accepting a harm in exchange for a benefit is a moral choice, and must be made openly by the people affected.
What could happen
Key architectural rules, failure boundaries…
Key architectural rules, failure boundaries, and verification steps for what could happen.
Inspect full technical derivation & primary sources ▾
- We want thisMost needs are met close to home
Requests are answered by people who already know each other, revealing almost nothing to strangers.
For example: a parent’s agent finds a babysitter among two families they already trust, without the request ever reaching the open network.
- We want to avoid thisThe search reveals too much
Requests leak details as they travel, and strangers piece together a profile.
For example: a request for a specialist doctor, passed through three rings, still carries enough detail for someone to work out who is ill.
- We want thisPeople find each other around shared goals
Agents introduce people who want the same thing, with consent, and projects form.
For example: twelve neighbours who each wanted a community garden are introduced by their agents and plant it the same season.
- We want to avoid thisRecognition becomes a new kind of inequality
Those already well connected collect most of the recognition.
For example: people with large trusted networks get most requests answered and most offers accepted, while newcomers wait.
What comes next
A simulation of the widening…
A simulation of the widening circles, measuring how fast answers arrive, how much is revealed and how easily the search is gamed. Simple, open formats for needs, offers and the records of help received. And a local memory for each person’s agent, kept on their own device, that they can export or delete in one step.