American Express is trying to solve a problem that arrives with autonomous shopping: how can a payment network distinguish what a customer asked an AI agent to buy from what the software actually attempted to purchase? TechCrunch reported on October 13 that the company has introduced an Intent Intelligence Framework designed to capture and authenticate a customer’s instructions before an agent completes a transaction.

The proposal is meant to carry those instructions through the transaction lifecycle rather than treating approval as a single moment at checkout. American Express hopes the framework will become an industry standard for agent-driven commerce, according to the report. It is being added to the company’s Agentic Commerce Experiences developer toolkit, which American Express launched in April with services for businesses building AI-enabled purchasing systems.

A secure abstract capsule holds budget, merchant and product-preference constraints.
The system captures purchasing constraints before creating a secure checkout credential.

The process begins with a Create Intent API. It records the customer’s purchasing request and any constraints attached to it, which could include the merchant, a maximum budget or product preferences. American Express told TechCrunch that the collection method may differ depending on the AI model, so the framework is intended to remain flexible enough to work across different agentic-commerce systems.

After the instructions are captured, the customer authenticates their identity and approves the recorded intent. The system then creates a unique Intent ID and a Proof of Intent Token containing the customer’s high-level constraints. That token is used to obtain a secure digital payment credential for the agent, allowing it to check out without using the customer’s actual card number while still tying the purchase to the authorized request.

That structure could let a payment provider intervene before an incorrect purchase is completed. TechCrunch described an example in which a customer requests a particular type and brand of shoe but the agent submits an order for a different brand. Because the approved constraints entered the transaction earlier, the mismatch could be detected and the payment stopped. The Intent ID and token could also provide evidence for fraud prevention and dispute resolution after a transaction.

An abstract AI reorders household supplies within a visible perimeter for price and quantity limits.
Standing permissions could support routine orders while preserving customer-defined boundaries.

American Express says it has not yet seen an increase in disputes caused by agentic commerce. Its decision is therefore a pre-emptive one, made as consumers show interest in delegating more shopping research to AI. In an American Express Trendex survey cited by TechCrunch, 66% of surveyed US consumers said they were likely to use an AI agent to research the best price for a product or service, while 61% said they were likely to use one for recommendations during the next 12 months.

The framework is also designed for a future in which an agent may not request fresh consent for every routine order. A customer might authorize repeated purchases of household supplies while setting limits on price or package size rather than specifying a merchant or brand. American Express argues that an automated check against the stored intent could provide an additional safeguard in those cases, while a record of which agent was authorized and when could clarify responsibility if fraud occurs.

The company is discussing implementation of its broader developer toolkit with partners across the payments ecosystem and says it will continue developing the system as agentic commerce changes. Adoption is not guaranteed, and the report does not establish that the framework will become an industry standard. But the proposal identifies an essential piece of infrastructure for autonomous shopping: a machine-readable boundary between delegation and permission, backed by an authenticated human decision.