The hardest problem in machine-to-machine payments is not moving value — stablecoins and HTTP-native protocols handle that well enough by now. It is establishing who, exactly, is paying. Alchemy’s AgentCard, built on its Mastercard integration, attacks that problem directly: each AI agent gets a dedicated email address, a phone number, and tokenized credentials, converting an anonymous piece of software into something that looks — to a merchant, a bank, a card network — like a counterparty it can underwrite. That is a quietly significant design choice, and it deserves a closer look at how the pieces fit together.
What AgentCard Actually Provisions
According to Forkast’s report, AgentCard treats autonomous software as a verifiable entity rather than a wallet with a bot attached. The dedicated email and phone number are not conveniences; they are the substrate that existing identity and verification systems already understand. A phone number can receive an OTP. An email can receive a receipt, a dispute notice, a refund confirmation. Tokenized credentials extend the same card-network tokenization model used for phones and browsers to non-human actors. In effect, the agent becomes addressable through the same channels a human customer is, which means merchants do not need to build new acceptance logic. The agent presents a tokenized credential where a card token would go, and the payment flows through infrastructure that already clears trillions of dollars a year.
Why Identity Precedes Payments at Scale
Protocols like x402 demonstrate that an agent can pay another endpoint in stablecoins over plain HTTP with no identity layer at all. That works for machine-to-machine micropayments between consenting software. It works far less well the moment a regulated counterparty is involved. A merchant acquirer wants to know who to charge back. A card network wants a token it can revoke. A fraud system wants device signals, history, reputation — none of which a fresh wallet address provides. AgentCard’s bet is that agent commerce scales not through pseudonymous crypto rails alone but through credentials that map onto the accountability structures of traditional payments. The dedicated email and phone are, in that sense, the agent’s paper trail.
How It Compares to the Stablecoin Agent Stack
The agentic-payments landscape is splitting into layers. On one side are settlement rails: x402, Coinbase Agent Payments, Skyfire and similar systems move USDC between machines. On the other side are identity and spend-control layers — AgentCard sits here, alongside wallet-guard approaches like the spend limits Ripple added to its XRPL agent kit. These are complements more than competitors: an agent could hold USDC for peer-to-machine settlement while presenting an AgentCard credential to a conventional merchant. The unresolved question is which layer captures the trust premium. Card networks charge for identity and dispute resolution; public chains settle cheaply but socialize fraud risk back to the transacting parties.
Open Questions
Three issues stand out. First, liability: when a credentialed agent buys the wrong thing at scale, the token can be revoked, but the dispute lands somewhere — issuer, agent operator, or the model vendor that made the decision. Second, revocation granularity: credentials are coarse instruments compared to programmable spending policies, and agents change behavior faster than humans do. Third, capture: if agent identity is issued through card-network integrations, the chokepoint for autonomous commerce sits with the same networks that currently sit atop human commerce — an outcome crypto-native builders explicitly set out to avoid. A neutral onchain identity registry remains an alternative, but none has the acceptance AgentCard inherits from Mastercard’s network.