Agentic Commerce: When Customers Send AI Agents to Shop

For as long as ecommerce has existed, we have designed for one kind of visitor: a person. A person who scrolls, compares, gets distracted, reads reviews, abandons a cart and comes back a few days later. Every conversion tactic, every landing page and every checkout flow assumes a human on the other side of the screen.

That assumption is starting to break. Customers are beginning to hand parts of the buying journey to AI agents. Find me the best option. Compare these three. Reorder what I bought last month. Book it when the price drops. The shopper is becoming partly software, and when that happens, much of what commerce teams have optimized for stops being what decides the sale.

From assistant to agent

It helps to be precise about what an agent is, because the word is used for almost everything right now. A system that answers questions about products is an assistant. Useful, but the customer still does the work. An agent is different. It can access context, use tools, remember what the customer wants, return to an unfinished task and actually act. It behaves less like a search box and more like a colleague doing the shopping on someone's behalf.

That difference is the whole point. The question for commerce leaders is not whether customers will chat with AI. It is what happens when software starts making and executing buying decisions inside limits the customer has set.

Discovery changes when the agent does the looking

Today, discovery is a fight for human attention: search ranking, visual merchandising, campaign creative, the hero banner. An agent does not look at your hero banner. It reads information, compares options, checks availability, price, delivery and returns, and filters on whatever its customer said matters.

That moves influence. If your product data is incomplete, inconsistent or locked inside images and marketing copy, an agent may simply not see you. If your delivery promise is vague, an agent cannot weigh it. The website as we know it may matter less, while the clarity of the information behind it matters much more. I sometimes call this the disappearing website. The interface is generated around the customer, and your brand has to show up inside it.

Voice adds another layer. I see voice as a natural interface because it shortens the distance between intention and action. "Order the usual" is a complete instruction to an agent. It is not a page view.

What loyalty means when an agent compares everything

An agent can compare every option, every time, without getting tired. That is great for customers and uncomfortable for brands that relied on habit or friction to keep people coming back.

I do not think loyalty disappears. I think it moves. It moves toward what an agent can verify and what a customer actually values: reliable delivery, fair and transparent pricing, easy returns, consistent quality and service that resolves problems. It also moves toward trust. The first AI agent your customer trusts may matter more than your next app, because that agent becomes the gatekeeper for many future decisions.

So the real question becomes: how does a brand earn the trust of the customer and stay understandable to the agent at the same time? Those are two different audiences, and most commerce teams are only designing for one of them.

Personalization without surrender

Agentic commerce makes personalization far more powerful. An agent that knows your size, your preferences and your budget can find things you would never have searched for. I find that genuinely exciting. Imagine a customer seeing clothing on a generated image of themselves before deciding to buy.

But personalization only works if trust survives it. Customers will not hand meaningful decisions to an agent that feels like it works for the retailer instead of for them. The direction I prefer is personalization that collects less, not more: clear value in return for data, real consent, customer control, processing on the customer's own device where that is possible, transparency about why something was recommended and the ability to undo it. I call it personalization without surrender.

Brands that build this way will be easier for both customers and their agents to trust. Brands that do not will find that the agent quietly routes around them.

Questions to take back to your commerce team

You do not need to rebuild everything this year. You do need to start asking different questions. Here is a short checklist I would bring to the next commerce, product and data meeting:

• Could an AI agent understand our products, prices, stock and delivery terms without seeing a single image?
• Which steps in our journey exist only because we assumed a human would be clicking?
• What would we want an agent to know about us that a comparison table would miss?
• How do our payment, returns and service processes work when the buyer acts through software?
• Where must a human stay involved because trust, emotion or risk is high?
• Which customer data could we stop collecting and still serve people well?

None of these require a big technology programme to answer. They require commerce, product, data, payments and service teams to look at the same journey from the point of view of a buyer who is partly software. Most of the work is in the operating model, not in the tools.

When the shopper is partly software, being understood matters as much as being seen.

Agentic commerce is one of the topics I am asked about most right now, by retail and ecommerce events, payments and marketplace audiences, consumer brands and financial services. The keynote maps the emerging agentic customer journey, the new friction points and growth opportunities, and where human trust still matters. I adapt it to your industry and your audience. If your event needs commerce leaders to leave with a clear view of what to prepare for, I would be happy to discuss bringing Agentic Commerce to your stage.