AI agents are getting closer to the moment when they can buy things for us. American Express wants to make sure they can also explain what we actually asked them to buy.
The company’s Amex Agentic Commerce Experiences Developer Kit includes an Intent Intelligence layer designed to capture, authenticate and carry a customer’s purchase instructions through the transaction. That could include the merchant, budget, product preferences and other limits before an agent is allowed to complete payment.
It is a small but important change in the logic of online shopping. The question is no longer only whether a payment is authorized. It is whether the purchase itself matches the customer’s intent.
AI Shopping Needs A Memory Of What The Customer Meant
American Express describes ACE as a suite of services covering agent registration, account enablement, intent intelligence, payment credentials and cart context. The intent layer creates an Intent ID and a Proof of Intent Token, while the payment layer gives verified agents tokenized credentials instead of exposing the customer’s actual card number.
That separation matters because large language models are capable of making plausible mistakes. In an example shared with TechCrunch, a customer might ask an agent to buy white Nike sneakers in size nine, only for the agent to submit an order for Adidas shoes. A conventional payment authorization could still approve the transaction. An intent-aware system could stop it earlier.
American Express says the same information could support fraud prevention and dispute resolution later in the transaction. Its official framework currently lists Account Enablement, Intent Intelligence and Payment Credentials specifications as available, while Agent Registration and Cart Context remain under development.
The company is also planning future protection for eligible purchases made by registered AI agents when authenticated intent has been passed to American Express. That protection is not simply a customer-service promise. It is an attempt to define who carries responsibility when an automated system acts incorrectly.
The Next Commerce Layer Is About Permission, Not Just Convenience
Agentic commerce is often presented as a faster version of shopping. But the harder problem is permission. A customer may authorize an agent to reorder household supplies with broad limits, while expecting much tighter control when the purchase involves a specific brand, price or category.
American Express says 66% of surveyed U.S. consumers are likely to use AI agents to research the best price for a product or service, while 61% are likely to use them for recommendations over the next year. Those numbers do not prove that consumers are ready to hand over the checkout button. They do show why payment companies are trying to define the rules before the behavior becomes routine.
This is also why trust signals will become part of the product experience. Google’s push to make AI content verification more visible reflects a similar shift: people need evidence about how an automated system produced an outcome, not just the outcome itself.
For retailers and brands, the implication is practical. Product data, pricing, inventory, fulfillment and returns will increasingly need to be legible not only to people and search engines, but to agents making decisions within customer-defined boundaries.
American Express is not just trying to make AI payments possible. It is trying to make customer intent portable across discovery, checkout, fraud checks and disputes. If that standard takes hold, the winning commerce systems will not be the ones that remove the customer from the transaction entirely. They will be the ones that let customers delegate without losing control.