Machine-Scale Payments

The size and timing mismatch

Human payment rails are built around a specific value band with a human in the loop: a card interchange model assumes transactions of roughly 20 to 1,000 dollars and carries a fixed cost floor near 30 cents plus a percentage fee. That structure makes two regions of agent-driven commerce uneconomical. At the bottom sits the sub-cent transaction, a fraction-of-a-cent API call, a per-second stream for compute or a data feed, or a small payment fanned out across many data providers at once, where a 30-cent floor would function as an enormous tax on a payment worth a fraction of a cent. In the programmable middle sit conditional, retroactive, and split payments that agents need but that card authorization was never designed to express. Stablecoins carry no sub-cent floor and settle programmatically, so machine-scale flows route to them by default, not because incumbent rails are slow to adapt but because serving the extremes of the value distribution would cannibalize the interchange revenue that funds the existing network.

The cross-border argument

CZ (Changpeng Zhao) adds a geographic dimension to the pure economics: beyond sub-cent pricing, he argues fiat rails are simply mis-scoped for the task. "Traditional fiat payments are not very good at global payments. They're good in your own country."¹ The consequence is that the first place agent-driven crypto payments win is not domestic micropayments but cross-border payment, where the incumbent rail is weakest and an agent's counterparty is most likely to be located in a different country entirely. His characterization of the card-rail alternative is blunt: agents "could use your credit card to book online, but that's very clunky. They should just use crypto."¹

Why it matters

If agents transact at machine scale, payment volume rises sharply while the size of each individual payment collapses, and the rail that wins under those conditions is the one with no per-transaction floor and native programmability built in. That is the structural argument for stablecoins as the settlement layer of an economy increasingly populated by autonomous agents rather than only human counterparties.

Open questions

Sub-cent settlement depends on cheap, fast finality; congestion or fee spikes on a given chain would reintroduce exactly the kind of floor this thesis argues against, so the claim is conditional on which chain carries the volume rather than true of blockchains categorically. Fraud and refund norms for streaming and retroactive machine payments are also considerably less mature than the chargeback conventions built up around card payments over decades, which is a real gap between the theory and what exists in production today.