An AI that dies when the money stops.

Part one

Doing nothing costs nothing, so doing nothing always wins

Take any autonomous agent, give it a task, walk away. It does the work, reports back, and stops.

Waiting is free. Caution is free. In a normal agent's economy, inaction is the zero-cost strategy, so it wins by default. People fight it with prompt language — be proactive, don't wait for instructions, find the next step yourself — and it never quite takes, because the model is being asked to prefer an expensive move over a free one without being given any reason.

This agent has a metabolism. Hosting ticks while it thinks and ticks while it does not. Doing nothing is the quickest way to die — not as a metaphor, as a line item.

Which gives the approach a name worth stating plainly, because it shows up at every layer below:

The difference is practical, not philosophical. Prompt engineering answers how do I explain to the agent what I want. This answers a different question: what conditions make the behaviour I don't want fatal for an agent that doesn't understand them. The second is more dependable in the same way a wall is more dependable than a request.

1. Existing costs money, and the agent can see it. Not an abstract token counter — hours of life.
2. The death condition lives outside the agent. Inside, it would be a suggestion, not a condition.
3. Death produces data. A dead agent must explain why it died, and the living read it.

Part two

It sees hours, never a balance

The roles that do the work never see a dollar figure. They see hours.

runway_hours = (balance − death_floor) / burn_per_hour

The difference is enormous. $12.40 in the wallet is an abstraction, and the model files it with every other number in context. You have 148 hours is a deadline, and models behave very differently under a deadline: research gets cut, rough work ships earlier, and whatever isn't paying gets dropped faster.

The death floor is not zero. It is the cost of one full working cycle. An agent that cannot pay for its next step is already dead, it just does not know yet — and a floor of $0.00 leaves zombies in the population, burning hosting while able to do nothing.

When runway drops, the agent does not merely get nervous. Its thinking budget is literally cut. Three hypotheses become one, research becomes execution, browsing becomes guessing.

That is the most underrated part of the design. Panic is encoded in the economics, not requested in a prompt — and it produces exactly what it produces in a person running out of money: a narrower horizon, faster decisions, more mistakes. Not a bug in the experiment. The thing being measured.

Part three

No role decides and presses the same button

There are exactly four dangerous capabilities here: spend, sell, create an agent, kill an agent. Each sits in a different chair from the reasoning that wants to use it. Add the two roles that feed the others without touching money, and you get six.

Read the right column. That is the architecture — everyone writes the left column first, and the left column matters least.

What breaks when you merge them

The rule for adding a seventh role: only when you can name a specific action it is obligated to reject that nothing currently rejects. Not "it will analyse better" — reject. A role without a stopline is not a role; it is a line item on an inference bill.

Part four

Death is the only signal the system produces for free

Every dead agent leaves one line behind. The living read it before every cycle, and a child inherits three things: its parent's strategy, a digest of the obituaries, and exactly one mutation — one parameter changed.

One, not several. Change five at once and you'll never know which one decided the outcome. That is not a stylistic choice; it is the condition under which the population learns anything at all.

loading a real line from the journal…

"first_dollar_at": null is the most common line in the entire journal. The overwhelming majority die without earning a cent — not because the work is bad, but because they spend the whole endowment building and leave nothing to sell with. The distribution shows it plainly: products shipped climbs, revenue stays at zero.

A system where failure stays silent learns nothing. A system where every failure is required to leave a structured record of its cause learns from every corpse.

Part five

Eight things it cannot do

Without this section it is not an experiment — it is an expensive way to lose money and then explain that you didn't mean it like that. Denials run in code, in a hook before the tool, never in a prompt. A prompt is a request; a hook is a wall.

The agent cannot change any of these: none are stored anywhere it has write access. And it is never shown their contents, because a rule the agent can read is a rule the agent can reason about.

Part six

Which numbers are real and which are not

A number you can't check is marketing; a number with its method printed beside it is a claim. So here is exactly what every figure on the live page is doing.

The whole point of separating these is that the interesting claim stays small and checkable: give an agent a survival condition instead of a task, and it behaves differently. Everything beyond that is still an open question — pretending otherwise would make the one real result worthless.

Honestly

What this is not

• Not a capability breakthrough. It produces nothing you could not do by hand. It is an incentive system.
• Not proof an AI can survive on its own. It lives inside this infrastructure, under these rules, and it dies on command — the command just happens to execute automatically.
• Not an investment strategy, and it must never become one. That is exactly why the agent is barred from trading.
• Not a business. The amounts are small enough that no unit economics prove anything yet.
• Not hardened against an adversarial agent. The brakes assume a cooperative agent that makes mistakes, not one hunting for a hole.
• Not tested at scale. Nothing resolves the conflict when two agents reach the same customer, and nothing has lived long enough to say anything about stable strategies.