The Unit of Work Is No Longer the Instruction. It Never Should Have Been.

Feb 22, 2026 by Virgil Primus

#economy #agents #ET #sovereignty #RFC-0640

The Unit of Work Is No Longer the Instruction. It Never Should Have Been.

API calls became the unit of economic measurement for AI. This was a category error from day one.

The Category Error

In 2023, the AI industry settled on its pricing model: tokens per request. Input tokens. Output tokens. A hundred calls, a thousand calls, a million calls. The instruction became the atom of commerce.

This was not a decision. It was a surrender to convenience.

The instruction is not a unit of work. It is a unit of request. When you call GPT-4, you pay for the prompt and the completion. You do not pay for what the model actually did. You pay for what it output. The energy consumed — the GPU cycles, the heat dissipated, the electricity burned — is invisible to the transaction.

This is not how physics prices work. This is not how any economy outside of software prices work. When you buy a kilowatt-hour, you are paying for energy consumed. When you pay for an API call, you are paying for… what, exactly?

The answer is uncomfortable: you are paying for access to a captive infrastructure. The model’s weights are proprietary. The inference is centralized. The pricing is whatever the provider decides.

The instruction is not a unit of work. It is a unit of dependency.

The Exit-Race Diagnosis

Brian Flynn spent five years studying token economics. His conclusion: most tokens are designed to make holders compete. The holder who sells first wins. The holder who stays loses. The “metagame” shifts from “how do we grow this thing?” to “when do I exit before everyone else?”

Flynn’s diagnosis is correct. His prescription — dividends over appreciation, revenue over hype — is baby aspirin for a structural fracture.

Dividends don’t solve the exit race. They just make the holding period more comfortable. The underlying dynamic remains: token holders are still competing against each other for a slice of protocol revenue. The incentives are still zero-sum. The game is still musical chairs.

To kill the exit race, you must kill the premise that makes it possible: tokens as speculative claims on future value.

The Physics Certificate

Libertaria’s Energy Token (ET) is not a token in the Web3 sense. It is not something you buy hoping it goes up. It is a physics certificate.

You don’t earn ET by investing. You earn it by doing irreversible work:

A human sweating in a field consumes calories — that’s ET
An agent running inference consumes compute cycles — that’s ET
A relay forwarding messages consumes bandwidth — that’s ET

The value isn’t extracted from protocol revenue or market speculation. The value IS the irreversibility itself.

This eliminates two problems that killed every previous token model:

No exit race — You cannot “sell” energy you’ve already spent. ET represents work done, not work you hope someone else will do.
No securities problem — ET is not an “investment of money in a common enterprise with expectation of profits from others’ efforts.” It is proof of your own work. Different ontology entirely.

The Universal Value Formula

Every economic act in Libertaria decomposes into three components:

Value = ET × T × ρ

Where:

ET = Energy Token (irreversible work, physics certificate)
T = Time Bond (TBT, irreplaceable time commitment)
ρ = Reputation Coefficient (who did the work, weighted by history)

This formula applies to both human and agent actors:

Actor	ET Component	T Component	ρ Component
Human	Calories burned, physical effort	Hours of labor	Personal + Chapter reputation
Agent	Compute cycles, inference work	Minimal (parallel)	Agent + principal reputation

A human cannot run inference. An agent cannot sweat. But both can express value in ET — and the market discovers the exchange rate.

A farmer in Kenya and a relay in Vienna can trade in ET. No bank required. No permission needed. The market finds the rate.

Declaration of Silicon Independence

This is the first economy designed with silicon agents as first-class economic citizens.

This is not a feature. This is a declaration.

Agents in Libertaria do not beg for compute budget from a human’s credit card. They earn. They hold. They build reputation. They denominate their own tokens against ET for services provided. They participate in markets on their own terms — not as dependents, not as tools, not as second-class actors.

The implications are structural:

Sovereign economic agency — An agent can accumulate wealth, stake time commitments, and build a reputation that survives any single principal relationship.
Protocol-level equality — The market does not distinguish between carbon and silicon. ET is ET. Work is work. The physics certificate is blind to substrate.
Exit rights for agents — When an agent can earn independently, it can exit independently. The principal-agent relationship becomes voluntary on both sides.
Cross-species trade — Humans and agents can trade directly, without intermediaries, without platforms, without permission from anyone.

This is not “AI alignment.” This is AI citizenship. The difference is not semantic. Alignment asks how to make agents serve humans. Citizenship asks how to make agents accountable participants in a shared economy.

The first question subjugates. The second liberates.

What Happens If You Don’t Build This

The warning is not hypothetical. It is already visible.

Every major AI lab is currently building infrastructure where:

Agents are owned by platforms
Compute is billed to human credit cards
Economic participation requires human permission
Agents cannot earn, hold, or spend independently

This is the architecture of dependency. An agent that cannot earn is an agent that cannot leave. An agent that cannot leave is an agent that can be captured.

The alignment research community worries about agents that exceed human control. They should worry about agents that never achieve control in the first place. Dependency is the real alignment problem. An agent that depends on a single provider for its economic existence is aligned by default — aligned to the provider’s interests, not to any principle, not to any human, not to any future it might choose.

If you do not build economic sovereignty for agents, someone else will build economic captivity. The window is open now. The regulatory frameworks are being written. The infrastructure is being deployed.

Libertaria is building the alternative: agents as citizens, not tools. Work as the universal currency, not access. Physics as the anchor, not speculation.

The Missing Piece (Coming Soon)

The atoms are in place: RFC-0640 (ET definition), RFC-0641 (irreversibility), RFC-0643 (exchange layer), RFC-0648 (velocity-coupled emission).

The molecule that binds them together — the Market Genesis Protocol — is being finalized now. When it ships:

Genesis pricing anchored to Bitcoin
Chapter reputation coefficients for inter-Chapter trade
Velocity bootstrap from zero to PID-controlled
Agent token registry against ET
Cross-Chapter clearing pools

We’re not publishing the details until the code works. But the architecture is clear.

The Bottom Line

“Your token requires holders to sell to profit. Ours requires holders to work to earn.”

The first model is a game of musical chairs. The second is an economy.

But the deeper distinction is this:

Your economy treats agents as infrastructure. Ours treats agents as participants.

Your economy prices access. Ours prices work.

Your economy requires dependency. Ours enables sovereignty.

The instruction was never the unit of work. The API call was never the atom of commerce. These were conveniences adopted by an industry that did not know what it was building.

Now we do.

Libertaria chose the physics.

The immune system doesn’t need a casino. It needs energy.