The integration is designed to provide AI agents a “complete identity and payment stack” and let them make online purchases on consumers’ behalf, according to a Thursday (June 18) press release.
“With AgentCard, an AI agent built on models from any provider, including OpenAI or Anthropic, can book a vacation, order groceries or renew a subscription on a consumer’s behalf, without the consumer ever touching a checkout screen,” the release said.
Setup takes developers less than a minute, giving agents everything they need to function, like a Visa payment token, a dedicated email address, a phone number, and a cryptocurrency wallet, via a single API, according to the release.
The launch comes a little more than two months after Alchemy introduced AgentPay, a tool designed to help AI payment systems work together.
The latest integration lets agents make transactions with Visa-issued tokens, “preserving rewards, credit lines and card benefits” without needing to create new accounts or credentials, the release said.
“Every major computing shift has produced a new kind of economic actor,” Alchemy Co-Founder and CEO Nikil Viswanathan said in the release. “The internet created online businesses. Mobile created the app economy. AI agents are next, and they need to be able to access the global economy, and AgentCard is how that starts.”
Tanner Riche, vice president for growth products and partnerships at Visa, said in the release: “As AI agents take on a more active role in commerce, they need to demonstrate trusted identity and seamless ways to transact. Visa is working with partners across the ecosystem so developers can bring these experiences to market on our Visa Intelligent Commerce infrastructure designed to support agentic commerce in a manner that is secure, reliable and ready to scale.”
Meanwhile, AI agents face a security problem, PYMNTS reported June 10. The internet cannot determine if an agent is authorized, who gave the authorization, and what limits it has.
“A person logs in once and leaves a behavioral trail fraud detection can read,” the report said. “Agents act continuously, spawn other agents and carry permissions that spread in ways no one mapped when access was first granted.”
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
