When AI agents shop for your customers: what to prepare, and what to wait on

Imagine a shopper typing one sentence to an AI assistant: find me running shoes under one hundred dollars, and buy them. The assistant compares stores, picks one and completes the purchase. No pages browsed, no buttons clicked by a human. This is called agentic commerce, and merchants are starting to ask what it means for their checkout.

What changes for a merchant

An AI agent reads product data, not pretty pages. If prices, sizes and stock are unclear or inconsistent, the agent may skip your shop and buy elsewhere. Then it reaches the checkout, and there may be no human to click the buttons at all.

So two things matter earlier than they used to: how clean your product data is, and how well your checkout works when the buyer is software acting for a person.

The standards are still competing

Several protocols for how agents discover products and pay have been announced:

• The Agentic Commerce Protocol (ACP), released openly in late September 2025.
• The Universal Commerce Protocol (UCP), developed for discovery through fulfillment.
• The Agent Payments Protocol (AP2), announced in September 2025 and donated to the FIDO Alliance on 28 April 2026.
• Separate agent-payment programs run by the card networks.

Coverage in 2026 describes the field as unsettled. One analysis argues the direction is shifting toward AI handling discovery while merchants keep control of checkout. Another advises merchants to plan for several protocols rather than bet on one. The main difference between proposals is who holds the payment credential at the moment of purchase.

Trust is the hard part

A survey cited in industry coverage found that 53.9% of consumers fear AI increases fraud risk. For merchants, that means each agent-made order needs to be easy to verify as real: who authorized it, for what, up to what amount. Until the standards settle, expect more review of these orders, not less.

What to prepare now

• Clean product data. Accurate prices, sizes, stock and descriptions, in a form a machine can read.
• A simple checkout. One that works when no human is clicking. Fewer steps, clear confirmation, clear refund terms.
• Clear order records. Who or what placed the order, and how it was authorized, so a dispute can be answered.
• Patience on standards. Watch how they settle before building around one.

What to avoid

• Do not rebuild your checkout around a protocol that may not win.
• Do not assume agent-made orders carry the same dispute risk as human-made ones. Nobody knows yet.
• Do not trust vendor claims about how much agent traffic you will get. There is no reliable benchmark for it.

Where we land

Agentic commerce is real enough to prepare for and uncertain enough not to bet on. The preparation, such as clean data and a simple checkout, helps with ordinary customers too.

The open question: would you let an AI agent complete a checkout without a human confirming? Automatic purchases are the point of the technology. Human confirmation is the safest guard against mistakes and fraud. As a shopper, and as a merchant accepting these orders, where do you draw that line?

EPaySe is a payment gateway in development for high-risk and regulated businesses. High-risk and regulated businesses are reviewed and approved case by case.

Sources:

• Transak, what is agentic commerce: https://transak.com/blog/what-is-agentic-commerce
• elogic, agentic commerce in 2026: https://elogic.co/blog/agentic-commerce/
• Checkout.com, OpenAI's agentic commerce shift: https://checkout.com/blog/openai-agentic-commerce-shift