You tell the agent what you want to buy
Take a headphone purchase as an illustrative example. You want to spend no more than €150, including delivery, and the headphones need to arrive by Friday. You would normally open several shops, compare models and read delivery terms. With agentic commerce, you can give that task to an AI agent.[S6]
The agent needs a request it can act on. In this example, finding a model is only part of the work: the budget includes delivery, and Friday is a deadline. Stripe describes a purchase flow in which an agent retrieves product and delivery options, presents them to the customer and submits a confirmed order.[S1] That is how the conversation can lead to an actual shopping task.
The agent needs to check the full price and delivery date
A well-reviewed pair of headphones at €139 sounds promising. Add €15 for delivery and it is over your budget. If it arrives next week, it does not solve your problem either. A product recommendation is only the beginning.
To meet your request, the agent needs the available model, the full price and the delivery date. Shopify documents catalog interfaces with product variants, availability signals and checkout links.[S2] Those details still need checking against the purchase you asked for. In our proposed example, a missing delivery date is a reason to ask or look for another offer, rather than promise that the headphones will arrive on time.
You decide whether the agent may place the order
In our example, the agent brings the finished offer back to you. You review the product, total price and delivery date, then approve the purchase. We chose this permission boundary for the example. What an agent is allowed to do needs to be agreed before it shops.
After your approval, the route depends on the shop. Shopify documents both a handoff to the merchant’s checkout and eligible direct completion by the agent.[S3] The supported flow and your permission determine which applies. In our example, look for the merchant’s order confirmation. A filled cart alone does not mean the headphones have been ordered.
Not every shop supports the whole purchase yet
For a retailer, this purchase has to work through a service that customers can actually use. Before choosing an integration, check which shopping steps that service supports in your market. An announced standard does not make your shop usable by every agent.
Google shows why the details matter. An integration needs Google’s approval before purchases can go live on its AI surfaces, and not every Universal Commerce Protocol, or UCP, feature is supported.[S4] According to the documentation we checked on 10 October 2026, the related integration hub in Merchant Center is initially rolling out for selected merchants and products sold in the United States.[S5] A retailer in Germany cannot treat that as confirmation that its own shop can participate.
Test a purchase from the request to the order confirmation
Start with one product and a shopping service your shop actually supports. Write down a customer request and what would count as a correct completed purchase. For the headphone example, that means the right model, no more than €150 including delivery, arrival by Friday, your approval and an order confirmation. These are chosen test conditions, not results we have measured.
Use that request to follow the purchase through the supported flow. Then repeat it with an unavailable variant or a changed price. Check whether the agent explains the problem and whether the customer can take over without starting again. Record where the journey breaks; that gives the team a specific next improvement to discuss.
Agentic commerce is a practical use of agent experience : people ask software to act on their behalf, and businesses need to make those tasks possible. MING helps design and test these journeys. Read more about our agent experience work .
