What is agentic commerce?

Agentic commerce is online buying and selling in which an AI agent acts on behalf of the buyer. The person states what they want. The agent finds options, compares them, and in many cases completes the purchase.

How it differs from ordinary e-commerce

In ordinary e-commerce a person visits your site, reads, decides and pays. In agentic commerce an agent visits your site, reads, decides within limits the person set, and pays or asks the person to approve. The merchant still sells the same product. What changes is who, or what, is reading the page.

The pieces involved

• The buyer, who sets the goal and the limits.
• The agent, which searches, compares and acts.
• The merchant, whose site and policies the agent must be able to read.
• The payment layer, which has to confirm that the agent is authorized.

Why merchants should care now

Choice is moving upstream. When an agent narrows ten options to one, the nine it skips never get a chance to make their case on the product page. The page still matters, but a machine-readable version of it matters first.

What agentic commerce is not

It is not a single product or a single company, and it does not mean people stop shopping. It is a change in the interface between a buyer and a store, and like earlier changes in that interface, it rewards merchants who are easy to understand.

A walk through one purchase

A concrete case makes the pieces easier to see. A person tells an agent: find a 500 ml olive oil, under a set price, delivered this week, from a store with a clear return policy. That one sentence contains a goal, a limit and two conditions.

The agent turns it into steps. It searches for candidates, opens product pages, and reads price, size, stock, delivery estimate and return terms. It drops the stores where it cannot find a required fact. It compares what is left. At the end it either buys within the limits it was given or comes back to the person with a short list and asks for a decision.

At each step the agent needs something from the store that a person would get by looking: a price it can trust, a size it can read, a delivery date, a policy. If any of those is hard to read, the agent has two options. It can guess, which risks a wrong purchase, or it can skip the store. Most agents are built to skip, because a wrong purchase costs more than a missed one.

Where it fails today

An agent can fail at any step of that walk. A price may be drawn as an image. A size selector may not say what is in stock. A cart may empty itself after a redirect. A checkout may require a step that only a person can do, such as a visual puzzle.

We catalog these failures in a shared taxonomy of seven classes, so that a store and an agent builder can describe the same failure in the same words. The taxonomy is on this site, and it is the basis of the scorecards we build.

The point for a merchant is plain. An agent does not leave a store because it dislikes the brand. It leaves because it could not finish. That is a problem a store can find and fix.

What to do first

Nothing here needs a rebuild. Three small steps cover most of the ground.

Make the facts readable. Price, availability, variants, shipping and returns should be present as text on the page, and in structured data that matches the page.

Keep the path short. Fewer steps between a product page and a completed order means fewer places for an agent to fail.

Test it. Run an agent through your own store and read the trace. A failure you can see is a failure you can fix.

Questions merchants ask

Does this mean people stop visiting my store?

No. People still browse, and many purchases will stay human-led. What changes is that a growing share of the first step, finding and comparing, may be done by an agent. A store that works for both audiences loses nothing.

Do I need to build something for agents?

Not a separate store. The same product facts that help an agent also help search engines and accessibility tools: clear text, consistent structured data and a short path to checkout. Start there.

How do I know if an agent can use my store?

Run one through it and read the trace. A scorecard that lists each task, whether it passed and where it stopped is more useful than a general impression.

Who is responsible if an agent buys the wrong thing?

That is still being settled. It depends on the limits the buyer set, what the agent was told and what the store showed. It is one reason clear product facts and clear authorization both matter.

Where Machine Commerce fits

Machine Commerce is an applied AI research lab for the agentic economy. We study how machines buy: how agents read product pages, compare offers and complete checkouts, and what merchants must change to be chosen.