AI agents look for payment rails

Polygon details its role in Agent Pay for Machines, where software agents, x402 and stablecoins meet.

Polygon Labs published on June 10 a useful clarification on Agent Pay for Machines, Mastercard’s new service for payments initiated by software agents. According to Polygon, Mastercard wants machine-to-machine transactions to be permissioned, orchestrated and settled at software speed across its global payments network, with more than 30 companies among the first participants or supporters. Polygon names Adyen, Ant International, Checkout.com, Cloudflare, Coinbase, Getnet by Santander, Global Payments, Solana and Stripe among them.

The most concrete signal concerns x402, an open protocol that lets AI agents pay for resources over HTTP, the basic protocol of the web. Polygon says that as of June 2026, 95% of x402 transactions settle on its network, representing more than 15.5 million agent payments to date. That should be read as ecosystem data from Polygon, not as an independent market measurement. Still, it shows that agent payments are no longer only a slide-deck concept. There is already measurable activity around automated microtransactions, often tied to compute, data, hosting or digital services.

The blockchain angle comes from the economics of these payments. A human payment is usually episodic: a person clicks, confirms and waits for a product or service. An agent can instead trigger a chain of tiny, continuous payments, for example to reserve freight, buy weather data, pay for compute or settle an API service. In that context, fees and finality become decisive. Polygon highlights finality of about five seconds and costs below one cent per transaction, as well as multi-rail settlement combining cards, accounts and stablecoins.

Caution is still needed. Agent Pay for Machines is presented as a phase for validating use cases and common rules, not as infrastructure that is already broadly deployed. The hard questions will include stablecoin compliance, verifiable agent identity, spending limits, liability when an agent makes a mistake and merchants’ ability to tell an authorized agent from abusive automation. Even so, the announcement clearly connects two movements: automation by AI agents and the return of stablecoins as a settlement tool. If software begins to buy and pay continuously, the value of a blockchain will not be judged only by throughput. It will also depend on its ability to prove who paid, with what authorization and under which limits.