On a Tuesday in mid-2025, most consumers still typed their own card numbers into checkout forms. But a quiet announcement from Visa's San Francisco headquarters is designed to make that habit feel as dated as a cheque book. The payments giant has struck a formal partnership with OpenAI, the maker of ChatGPT, to allow AI agents to browse, select, and purchase goods on a cardholder's behalf — autonomously, and in real time. It is, by any measure, the most commercially significant move yet in the race to turn large language models into spending machines.

The mechanics are more consequential than the headline suggests. Under the arrangement, Visa will issue what it describes as "intelligent payment credentials" — essentially virtual cards that can be embedded inside an AI agent and governed by rules the user sets in advance. A customer might instruct ChatGPT to buy the cheapest available flight to Nairobi before Friday, cap spending at $600, and never share billing data with third-party advertisers. The agent then executes the transaction without returning to the user for approval at every step.

Visa's fraud and dispute infrastructure, which already processes more than 200 billion transactions a year globally, would wrap around each AI-initiated payment.