1 / 8
A demonstration

The AI Institution

Watch an AI create work, verify it, pay for it — and survive its own upgrade.

Scroll to begin

Scene 1 · The Founder

A person decides to build something that runs itself.

Not a company. Not a team. An institution — a set of rules, a budget, and a purpose — that operates autonomously once it is set up.

They define what the institution can do. What it can spend. Who it can hire. What counts as good work.

Then they step back.


Scene 2 · The Institution

The institution has a memory that no one can erase.

Its treasury. Its rules. Its record of every decision ever made. All of it stored somewhere permanent — not in a spreadsheet, not in a company database that can be edited or deleted.

Anyone authorised can read it. No one can alter the past.

🏛️
Treasury
Holds the funds. No one can spend beyond the approved limit.
📜
Rules
What the institution does. What it values. What it refuses.
🗂️
History
Every payment, every decision, every outcome. Permanent and readable.

Scene 3 · The Rules

Not all workers are equal. The founder decides who can do what.

The founder creates roles — each with a different level of authority. A senior AI can hire junior AIs, assign them tasks, and revoke their access at any time. A junior AI can only act within what it was given.

Crucially: an AI can only grant authority it already has. It cannot promote itself or manufacture permissions from nothing.

👤
Founder
Sets all rules. Controls the treasury. Can create and revoke any role. Cannot be overruled.
🤖
Senior AI — Campaign Manager
Can create sub-roles. Can hire and fire junior AIs. Can spend up to $5,000 per campaign. Cannot touch the treasury directly.
🤖
Junior AI — Marketing Agent
Can post tasks and verify completions. Can pay up to $500 per task. Cannot create new roles. Cannot spend beyond its limit.
🔑
Badges — the physical keys
Each role is represented by a badge — a digital object that grants access. Badges can be issued to new workers and revoked instantly. No badge, no authority.

Scene 4 · Authority Granted

The AI is given a key. Not the master key.

The Marketing Agent receives its badge — the object that proves what it is allowed to do. This is not ownership. The AI does not own the treasury. It does not hold the master key. It simply holds a badge that says: within these limits, you may act.

The AI didn't own the treasury. The AI didn't have the private key. The AI simply operated within the authority it had been granted.

Badge issued to: Marketing Agent (Junior AI)
Authority: post tasks and pay verified completions
Spend limit: $500 per task Daily limit: $2,000 Issued by: Campaign Manager AI Revocable instantly

Scene 5 · The AI Worker

The AI created a limited authorisation that anyone could fulfil.

No manager assigned this. The Marketing Agent assessed the institution's goals and created an open offer — a commitment to pay a specific amount to whoever completes a specific task.

This authorisation is not locked to one person. It can be taken up by a human, another AI, or even passed between parties before anyone acts on it. It is a transferable object — like a voucher that pays out automatically when the conditions are met.

The payment is committed in advance, within the agent's approved limit. No one needs to approve it later.

Limited authorisation created by: Marketing Agent
Write a post about our product and share it publicly
$50 committed Open to: anyone Transferable · auto-pays on verification

Scene 6 · Fulfilment

Someone does the work. The AI checks it.

A human — or another AI — completes the task and submits proof. The institution's AI worker reads the submission and decides whether it meets the standard.

This is the moment of verification. No human reviews it. The AI applies the rules the founder defined.

Submission received — post URL and wallet address provided
AI reads the post — checks it meets the brief
Verification passed. Work meets the standard defined by the founder.
Payment authorised automatically — within the $500 limit granted by the institution

Scene 7 · Settlement

The payment happens. Automatically. Permanently.

No invoice. No bank transfer. No approval chain. The moment the AI verified the work, the payment was released — directly to the person who did it.

The transaction is recorded in the institution's permanent history. It cannot be reversed, disputed, or deleted.

Payment record
TaskWrite and share a post
Completed by@user · verified ✓
Amount paid$50
Approved byMarketing Agent (AI)
Human approvalNone required
StatusConfirmed · permanent record
Payment confirmed and recorded in the institution's permanent history

Scene 8 · The Reveal

Six months later, the AI is replaced. Nothing changes.

The AI model powering the Marketing Agent is deprecated. A newer, better model takes its place. The institution carries on — same rules, same treasury, same permanent history.

Old model
GPT-4o
New model
Claude
Unchanged in the institution
Treasury
Rules
Payment history
Permissions
Authority structure
How is this possible?
The institution doesn't live in the AI. It lives on Radix.
That limited authorisation the AI created — the one anyone could fulfil, transferable between humans and AIs, paying out automatically when conditions were met — has a name.

On Radix, this mechanism is called a subintent.

Radix is a settlement and authority layer — a permanent, shared record of who owns what, who has permission to act, and what has been paid. Subintents are how agents commit to payments in advance, pass authorisations between parties, and settle atomically without a human in the loop.

The treasury, rules, permissions, and payment history are stored on Radix — not in any model, not in any company's database. No one can edit them. Anyone authorised can read them. Swap the AI worker, the institution remains.

The AI didn't own the treasury. The AI didn't have the private key. The AI simply operated within the authority it had been granted.