future of e-commerce
Agents shop for people. How to prepare your store for a world where the customer never visits your site
AI assistants are no longer just recommending products. They’re starting to compare them, choose them and complete the purchase. Here’s what that changes for a store we designed for people.
For twenty years, we designed online stores for people. Page layout, photos, button color, the order of arguments on a product page: all so that a person who landed on the site would want to stay and buy. That model isn’t going away, but a second one is growing next to it, where nobody visits the site at all. An agent does.
An agent is an AI assistant that gets a task from a person, not a query. “Find me a gray bed for my greyhound for the living room, under PLN 400, with a washable cover, delivered by Friday.” The agent compares offers across several stores, checks availability and delivery times, rejects whatever doesn’t meet the conditions, and presents two or three options. In more and more cases, it can also complete the purchase. The person only sees the result.
What an agent sees when it looks at a store
An agent doesn’t see the lifestyle photo of a dog on a rug. It doesn’t feel the tone of the description. It doesn’t notice the button is yellow. It sees data: whether the product has features it can filter by, whether price and availability are up to date, whether dimensions are given in units it understands, whether the return policy is written down somewhere in a readable form, whether the store even responds fast enough. A store that was beautiful for people can be a blank page to an agent.
That flips a few habits. A description that “sells with emotion” still matters to the person who approves the choice at the end, but it won’t get past the agent if there are no facts under the emotion. A photo without alt text is invisible to an agent. A promotion announced in a banner instead of in the price data doesn’t exist. Variants that appear in the store as three separate products look to an agent like three inconsistent offers.
Three things a store needs for an agent to choose it
- Complete, up-to-date product data in a standard format. It’s the same foundation we write about when we talk about order and GEO. An agent is simply the most demanding reader of that data.
- Clear terms: price including shipping, lead time, returns, warranty. An agent works within the constraints its person gave it, so it will reject an offer where it can’t verify a condition.
- Trust signals that can be verified: reviews, number of units sold, information about the manufacturer and brand, consistent data across sources. An agent choosing between two similar products will pick the one it has fewer doubts about.
What this means for marketing
Part of the budget that today goes into bringing people to the site stops making sense in an agent-driven model, because the person doesn’t come. What grows in importance is everything that builds a brand’s position outside the site: reviews, presence in comparisons, expert content an agent can cite to justify its choice. A brand people write about with “I bought it and now my dog won’t sleep anywhere else” has an edge no campaign can buy.
This isn’t the end of campaigns or of beautiful stores. It means a store has to work on two layers at once: for the person and for the machine helping them. The good news is that both layers feed on the same clean data. The bad news is that stores without it will gradually drop out of the buying process without any dramatic decline showing up in their stats. Fewer and fewer people will simply find them, and nobody will know why.
Where to start
With a test anyone can do in fifteen minutes. Ask an AI assistant to choose a product for you from your own category, giving it a few specific conditions. See whether your store even comes up, and if it does, with what data. Then check which information the agent couldn’t find. That’s your to-do list for the next quarter.
