The Intelligence Economy Needs an Accounting Layer

When we talk about an "intelligence economy," the conversation usually fixates on capability — what models can do — and rarely on plumbing. Yet every durable economic transformation in history has been underwritten by an unglamorous infrastructure layer: double-entry bookkeeping for merchant capitalism, clearing houses for industrial finance, the container for global trade. The age of artificial intelligence will need its own.

The problem is legibility, not capability

Autonomous agents are already beginning to do economically meaningful things: provisioning compute, paying for data, sub-contracting tasks to other agents, and committing their principals to obligations. The technical capacity to act is no longer the binding constraint. The binding constraint is legibility — whether a human institution can, after the fact, reconstruct what happened, who authorized it, and what it cost.

An economy that cannot be audited cannot be trusted, and an economy that cannot be trusted cannot scale.

Today, most agentic activity is logged as application telemetry: useful for debugging, useless for accounting. There is no shared notion of an account, a ledger entry, or a counterparty that an auditor, a regulator, or a finance team could rely on.

Three properties an accounting layer must have

We argue that a credible accounting layer for autonomous economic activity needs three properties, in order of difficulty:

1. Identity that survives delegation. When agent A acts on behalf of principal P and sub-delegates to agent B, the chain of authority must be recoverable — not inferred from logs, but recorded as a first-class fact.
2. Settlement with finality. Obligations between agents must resolve to states that both sides, and a third-party observer, agree are final. This is where tokenized settlement rails become relevant — not as speculation, but as plumbing.
3. Disclosure that maps to existing standards. The output has to land in the vocabulary that finance and audit already use. A novel ledger that no controller can sign off on is a research artifact, not infrastructure.

What this is not

This is not a call for a single global ledger, nor for putting every agent interaction "on-chain." Most activity will remain off-ledger and that is appropriate. The argument is narrower: the economically consequential fraction of agentic activity — the part that moves value or creates obligations — needs a settlement and accounting substrate designed for it, rather than retrofitted from observability tooling.

The Intelligence Economy Institute's first body of work is to specify what that substrate looks like, and to convene the finance institutions, auditors, and standards bodies who will ultimately have to accept it. The capability is arriving on its own schedule. The infrastructure will not.