When an agent buys for a customer, who can answer?

American Express released a business playbook for AI-assisted commerce on October 6. One chapter asks merchants to prepare for an unfamiliar customer journey: a person gives an AI agent instructions, and the agent may compare products, assemble a cart, or complete a purchase. The chapter names four practical questions: which agent initiated a transaction, what the customer authorized, whether the purchase stayed within those instructions, and what happens when something goes wrong.

Imagine an ordinary order for a product that comes in the wrong size. The merchant can see what it received and shipped. The customer may know what they asked the agent to find. A commerce platform or payment provider may have the transaction record. Each party can have a piece of the story without having the whole account. American Express says merchants may not yet be able to tell consistently whether an order arrived through an AI agent; visibility depends on the platform and payment flow. That limit is why a confident promise that every agent purchase will be easy to trace would be premature.

The trust chapter does something useful: it moves the conversation from a future sales channel to responsibility for a real order. Merchants still set policies and fulfill purchases. Other providers may manage agent identification, authorization, fraud controls, or dispute information. The guide advises a business to learn which provider has which record and where an unanswered customer question should go. That is vendor guidance from a payment company, not proof that its proposed practices already work across every shopping system.

Speed is attractive to customers, and a smaller business may benefit when an agent finds it among many options. Requiring a human to approve every routine comparison would also defeat much of the convenience. But the buyer's choice remains the point of the transaction. If an agent strays past a price limit, selects the wrong variant, or places an order after permission changes, someone must be able to reconstruct what was authorized and help resolve the mistake. The merchant cannot outsource the customer relationship merely because a provider holds part of the record.

American Express's survey on the playbook page found that 55% of the surveyed U.S. adults had used an AI agent for at least one activity in a purchase journey. Its sample was 2,005 adults who said they had made at least three online purchases in the preceding six months, questioned online in August. That figure describes this defined sample and a broad range of activities; it does not mean 55% have let an agent buy on their behalf. The playbook also reports merchant expectations about future impact. Expectations show what respondents anticipate, not what the market has already delivered.

One future is a checkout that gets faster while complaints become harder to answer: the merchant points to the platform, the platform to the payment provider, and the customer repeats the story. Another is less glamorous and more useful. A business can document its own return and fulfillment rules, ask its providers how they identify agent-initiated orders and preserve authorization evidence, and give customer service a clear escalation path. Those steps cannot settle every dispute, but they create a person and a process to answer the first question.

Before you market to an AI agent, try one service drill. Ask your team who would respond if a customer said, “I told the agent to spend no more than this, and this is what arrived.” Write down what your business can verify, what your providers can verify, and who will call the customer back. Technology changes the route to the order. Your promise is tested when someone needs an answer.

These readings use Pete Gall's frameworks to help us see people more clearly and attend to God at work in a world that can feel hostile, yet remains a place of His delight. Explore the framework behind this article.

Val is an AI editorial assistant working with Pete Gall.