The value of AI is never just the output. Well curated, it is the quality of the decision it enables.
The Village is a private online home that a community owns outright. This piece opens with how it is built, stated in full technical terms — for the reader who wants the engineering before the argument. Everything in that opening is then explained in plain language through the rest of the piece; if it is not your register, read straight past it and lose nothing.
The record. Records and decisions are written as append-only, tenant-scoped entries to a per-community ledger, and sealed with tamper-evident Ed25519 receipts over a canonical form. The signing key is published and resolved by did:web, so a seal can be checked against a key the community controls. A standalone verifier — no platform code, no network call — recomputes the canonical hash and checks the signature offline; tampering, reordering or substitution fails. The layer above this, which carries a binding governance verdict rather than an integrity seal alone, is in development.
Isolation. Queries are scoped by tenant by default and fail closed — with no tenant context it returns nothing. Community data does not cross the tenant boundary. Administrative access to a community’s contents is treated as a security defect, not a convenience the platform reserves.
Intelligence. Inference runs on self-hosted GPUs in New Zealand and the EU, on open-weight models, retrieved and grounded against the community’s own corpus. No inference is sent to a US-owned cloud.
Synthesis. Deliberation output is generated under hard invariants enforced in code, not by policy. A synthesis that would express a position as a mean, a score, a tally or a percentage is refused before it is stored; a synthesis that would drop a recorded minority or values objection is refused. Objections are typed — safety, values, effectiveness, process, resources, other — and carried by name into the result. The failure mode is refusal, not silent averaging.
Federation. Reconciling many rooms into a single account runs the same invariants — no cross-room averaging step, the same refusal on failure. It is built and runs today, but under a per-community switch that is off by default: a rehearsal capability, not yet a standing service.
Ground. Hosting is on European (OVH) and New Zealand (Catalyst) infrastructure, under EU and New Zealand law, with no US-owned dependency in the serving path.
In development. One substrate provisions many kinds of Village today. The agentic layers above it — tools that act, and infrastructure that makes and supports Villages with less human labour — are still being built; we mark them as direction, not as finished work.
Every community now lives somewhere online, and almost none of them own the ground they stand on. Their records, their conversations, their decisions rest on platforms held by companies that can change the rules, read the contents, or sell the lot — and that reserve, in writing, the right to do all three. This was a poor bargain when the platform was only a landlord. It became a dangerous one the moment a community’s own memory and judgement turned into the very material a new kind of software is built to rewrite.
The Village is our answer to that bargain: a private home for a community, built so that the community owns what happens inside it. It runs on machines we hold in New Zealand and the EU. Beneath it lies a record the community keeps itself, sealed so that any later tampering shows. The point is not that we promise to behave; it is that a member can lift a decision out of the system, carry it to their own computer, and confirm it is untouched without asking us anything at all.
Four pillars keep a Village in its members’ hands.
Its own record — the community’s memory, kept so that any tampering shows, and checkable by any member alone.
Its own intelligence — models that run on our machines and learn from the community’s own material: of the community, not a foreign model serving someone else.
Its own tools — the software is ours to run; no large supplier sits beneath it to revoke it, meter it, or watch it.
Its own ground — it answers to New Zealand and European law, not another jurisdiction’s terms of service.
A Village is a group of people with something in common — a marae, a club, a congregation, a small business, a neighbourhood. It is not a crowd and it is not a market. It gathers people around a shared purpose and gives them the means to act on it together, and it lets them join with other Villages when it suits them without being folded into one. The instinct to belong to a group of workable size is old and will not be argued away; the Village takes that instinct as it is and gives it good ground to stand on.
The clearest way to see what a Village is for is to watch one make a decision.
An Assembly is a Village deciding something that matters. Its members put up proposals, argue them, and set down where they stand; the disagreement is recorded, not settled by weight of numbers. So far this is only good process. What turns it into something new is what happens next — and it is not email, nor a message thread, nor a poll with a winning line. The deliberation of a whole community — hundreds of arguments, positions and reservations — is read and drawn together by models held to a single unbreakable rule: they may find the shape of a room’s thinking, but they may not average its people away. A minority position is not shrunk to a percentage. It survives into the result, named, with its reasoning attached.
That is hard at the scale of one Village. The harder and more consequential thing is to do it across many. When Assemblies join — a hundred rooms, a thousand — the same machinery reconciles what each has decided into one legible account of where a whole population stands, and why, without collapsing it into a yes or a no. This is the reach the old tools never had. A referendum crushes a nation into a single bit. A survey grades a set of questions someone else chose. Neither can carry a society’s real reasoning — its breadth, its dissent, the texture of why people hold what they hold — from the room to the scale of the whole. This machinery can, and that is what it is for.
The Assembly is built and working; the machinery that joins rooms is built too, and runs under a per-community switch that is off by default. The architecture is the same at every size — the single room and the society-wide result are one thing at different scales. Getting to the largest of them is a matter of people taking it up.
The four pillars are usually sold as reassurances — privacy, ownership, independence. They are more than that: they are the conditions under which it becomes safe to let a machine act on a community’s behalf.
An agent is only as trustworthy as the ground beneath it. Give it a memory it can quietly alter and it can rewrite the past; ground it in a model answerable elsewhere and it serves elsewhere; build it on tools a supplier can change under you and it is the supplier’s to redirect; place it under a foreign jurisdiction and the law that reaches it is not yours. Take those four risks away — a record that cannot be altered unseen, a model of the community, tools the community holds, ground under its own law — and only then is it safe to let the machine act.
That is the turn from a Village to agentic infrastructure. On this ground a model can be trusted to do more than answer — to act within the community’s own rules, on a record it cannot forge, under law that is the community’s own. And what one Village’s agent does for a community, a federation of them can do for many — the deliberation a single Assembly holds becomes, across many rooms, a reasoned account of where a whole society stands, its dissent intact.
This is the doorway Harari points at from the far side. His warning is that AI is learning to master the systems that hold a society together — its records, its institutions, its shared sense of what is true — and that whoever masters those systems, in the end, governs the society itself.1 The Village is that same architecture answered from the inside. A community that keeps its own record, runs its own intelligence, holds its own tools and stands on its own ground can meet that future as a participant, not a subject. None of this slows what AI does in the world; it settles who a community meets it as. The four pillars are why its agents can be trusted; the agentic infrastructure is the path they open. Where that path leads — communities co-operating at the scale of a society without surrendering what makes each of them their own — is still ahead of us, and we name it as a direction, not a thing already delivered.
AI will have more ideas than people ever could — but it takes people to understand where they lead.