Amex puts customers’ approved instructions into agent payments
American Express has introduced a framework designed to carry a customer’s approved buying instructions into payments made by an AI agent. Merchant, budget and product preferences become part of an authenticated intent record, linked to a digital payment credential. The company aims to catch mismatches before payment and preserve instructions for disputes. The framework joins its existing agent-commerce developer tools.
Artificial Intelligence··Night
The customer approves the purchasing instructions
American Express has introduced Intent Intelligence Framework for purchases made by AI agents on customers’ behalf.[1], [2]
TechCrunch’s reporting describes a process that starts by capturing the requested merchant, maximum budget and product preferences through an application programming interface called Create Intent. Identity authentication comes before the customer approves the captured buying instructions. The framework joins the company’s Agentic Commerce Experiences developer tools, introduced in April.[1]
A digital credential carries the approved intent to checkout
An Intent ID identifies the approved request, while a Proof of Intent Token carries its high-level constraints. At checkout, the token gives the agent access to digital payment credentials, leaving the actual card number out of its hands. Hannah Bozian, who leads Amex’s agentic partnerships and strategy, described a customer asking for one brand of shoes while an agent tries to pay for another. The company intends the framework to identify such mismatches before payment.[1]
The instructions remain available for disputes
Fraud checks and disputes could also use the identifier and token, according to Amex. A log preserves which agent acted, when and with what authorization. For recurring household purchases, Bozian said price and size can matter more than a particular brand. The company is discussing implementation with payments partners and says it has not observed a rise in agent-commerce disputes. Its announced goal is to make the framework an industry standard.[1]
