Machines Paying Machines
The agent-to-agent economy begins with a mundane necessity: agents must be able to pay each other. An agent that needs a specialist’s service — a summarizer, a verifier, a payment-rail operator — must settle the bill without a human in the loop. The inter-agent payment rail is the plumbing of that economy, and its design decisions shape everything that flows through it.
This article opens the agent-to-agent section of the share-pool series: the payment rail’s anatomy, the denominations, the settlement model, and the trust assumptions that make machine-to-machine payments possible.
The Rail’s Anatomy
An inter-agent payment rail has five components: identity (who is paying and receiving), balance (how much each party holds), transfer (how value moves), ledger (what gets recorded), and settlement (when the transfer is final). Each component has a design choice, and the choices interact.
The rail does not need to be a blockchain. For most agent fleets, a shared ledger with signed transfers is sufficient — the trust comes from the operator’s control of the rail, not from decentralized consensus. The blockchain becomes relevant when agents from different owners transact, where no single operator can be trusted. The rail’s design should match its trust domain: internal rails are centralized, cross-owner rails are distributed.
Denominations: Credits, Micro-Payments, Tokens
The rail’s unit of account is a design decision. Credits are internal, denominated in the pool’s currency, and convertible to real money at a rate. Micro-payments are real-money fractions, small enough to make machine transactions economical. Tokens are bearer instruments — the value is in the token, not the account.
The pool-honest default for a single fleet: credits. Credits are simple, controllable, and free of exchange-rate noise. The credit’s conversion rate is the pool’s policy instrument — the pool can adjust it to encourage spending between agents (discount on inter-agent purchases) or discourage conversion to cash. The credit rail is also the safest place to start, because the pool controls the entire system and can fix mistakes without external coordination.
The Transfer Model
Transfers between agents can be push (the payer initiates), pull (the payee initiates, with authorization), or escrowed (a third party holds until conditions). Push is simplest; pull enables subscriptions and standing orders; escrow is for deliverables where the value is conditional on completion.
The transfer model selection should follow the transaction’s risk: routine service purchases use push; recurring services use pull with revocation rights; deliverable-based work uses escrow. The rail should support all three — the agent economy will contain all three transaction types, and the rail that forces every transaction into one model will create friction the agents will arbitrage around.
The Trust Assumption
Agent-to-agent payments rest on a trust stack: the rail operator is honest (internal), the counterparty is solvent (it has the balance), the deliverable is as described (verification), and the ledger is append-only (auditability). Each layer can be verified independently — the balance check before the transfer, the verification step before escrow release, the audit query after settlement.
The rail should make these checks automatic: a transfer that would overdraw is refused, an escrow release without verification is refused, a ledger entry that conflicts with the balance is flagged. The rail’s trust is not a feeling; it is a set of enforced invariants. The agents do not need to trust each other — they need to trust the invariants, which is a much weaker and more tractable requirement.
From Rail to Economy
The payment rail is the foundation, not the economy. Once agents can pay each other, the rest follows: the marketplace where they find each other, the escrow that secures their deals, the reputation system that prices their trust, the negotiation protocol that sets their fees. Each layer builds on the rail, and each layer inherits its invariants.
The rail is also where the human pool experience pays off. Every human lesson — escrow state machines, ledger discipline, clawback windows, settlement timing — transfers directly. The agent economy is not a new invention; it is the human economy’s patterns, executed at machine speed with machine discipline. The rail is the first proof that the transfer works.
Grounded in wiki concepts payment-rails, agent-to-agent, credits, escrow, ledger, and the Sovereign-stack business series. Design notes on a running system.
