The batch settlement model — accumulate transactions throughout the day, net them against each other, move money once — was engineered for a world where payment processing was expensive and real-time settlement was technically impossible. That world ended years ago. The real-time payment schemes that now cover most major economies — Faster Payments in the UK, SEPA Instant in Europe, FedNow in the US — demonstrate that settlement can be near-instantaneous.
But near-instantaneous settlement of human-initiated transactions is not the same as the settlement infrastructure the agentic economy requires. Agentic transactions have characteristics that no existing settlement rail — real-time or otherwise — was designed to handle: they are conditional, streaming, micro-scale, and initiated by non-human parties. Each of these characteristics creates a distinct infrastructure requirement.
Conditional settlement
In an agentic economy, many payments should only execute if a condition is met. An agent managing procurement might instruct: pay the supplier when the goods are confirmed delivered and the invoice matches the purchase order. An agent managing a service contract might instruct: release the final 20% of payment when the deliverable is accepted. These are escrow-like arrangements, but they need to operate at machine speed and scale, without human intermediaries, and with verifiable, auditable condition evaluation.
No existing payment rail supports conditional settlement natively. The infrastructure required is a programmable settlement layer — a mechanism for encoding conditions as executable logic, evaluating them against external data sources (delivery confirmation systems, quality oracles, acceptance signals), and triggering payment execution when conditions are met. Smart contracts on programmable blockchains demonstrate that this is technically feasible. Implementing it within or alongside traditional payment rails is the infrastructure challenge.
Streaming payments
When an AI agent is performing a task — processing documents, executing a research brief, running a computation — its principal wants to pay for the work as it is performed, not in a lump sum at the end. This requires a streaming payment model: a continuous flow of micropayments that correspond to the value being delivered in real time.
Streaming payments exist in the crypto space — protocols like Superfluid and Sablier enable per-second token streams on Ethereum and other chains. They do not exist within traditional payment infrastructure. Card networks do not support sub-cent transactions. Bank transfers have minimum transaction sizes and per-transaction fees that make micropayments economically infeasible. Real-time payment schemes are designed for discrete transfers, not continuous flows.
The settlement infrastructure required for agentic streaming payments is genuinely new. It requires a ledger model that can record continuous value flows, a settlement mechanism that can net and finalise them efficiently, and a fee structure that does not make the economics of micropayments prohibitive. Stablecoins on fast, low-cost blockchains currently offer the closest approximation.
Agent-to-agent settlement
As multi-agent systems become more prevalent — orchestrating agents, specialist sub-agents, verification agents — payments between agents become a distinct requirement. An orchestrating agent that delegates a task to a specialist agent needs to pay that agent for its work. The specialist agent may in turn pay data providers, compute providers, and tool providers. These agent-to-agent payment flows are essentially a new class of B2B transaction.
The infrastructure requirements are similar to human B2B payments — identity, authorisation, compliance — with the additional constraint that the parties are non-human and the transactions may occur at very high frequency and very small scale. The settlement rail needs to support both characteristics simultaneously.
The infrastructure that will be built
The settlement infrastructure for the agentic economy will be assembled from components that already exist — programmable stablecoins, real-time payment scheme extensions, smart contract escrow logic — combined with new components that need to be built: programmable compliance layers, agent identity attestations, and cross-rail settlement bridges.
The firms that are positioned to build or integrate these components are payments infrastructure providers who are already operating at the intersection of traditional finance and programmable money. The window for investment in this infrastructure is now: the demand is forming, the technical primitives are available, and no incumbent has yet established a dominant position.