For twenty years buying online looked the same: a person searches, compares tabs, and clicks buy. More of that work is now handed to an AI assistant. The person says what they want; the assistant searches, compares, and increasingly completes the purchase. That is agentic commerce.
Definition
Agentic commerce is commerce carried out by AI agents acting for buyers or sellers.
A buying agent acts for the customer: "a carry-on suitcase under 150 euros, delivered this week, from a shop with free returns." A selling agent acts for the merchant: it answers questions, recommends the right product, and closes the sale.
The difference from classic e-commerce is who is on the other side of the interface. Storefronts were designed for people: images, filters, buttons. An agent does not look at the page. It calls interfaces and reads structured data.
Being visible means being callable
Assistants are gaining tools, memory, and the ability to pay. Protocols such as MCP standardise how an agent discovers and calls external systems.
Today, being visible means ranking on a results page a person reads. As agents do more of the buying, it also means being callable: your catalogue is one of the places the agent can query, with data it can parse and operations it can run.
What "agent-ready" means
A business is agent-ready when its catalogue, stock, prices, and transactions are available through structured interfaces an agent can call. In practice:
- Structured output. Machine-readable data with a stable schema: product, price, availability, variants, delivery terms.
- Search and detail. The agent can search the catalogue with filters and fetch full detail for one item.
- Transactions within limits. The agent can build a cart and move towards checkout, within rules you set.
- Discovery. The interface is described so agents can find and understand it, for example through an MCP server.
A chat widget on your website is not agent-ready. The test: can an agent, given a goal and a budget, complete a purchase against your systems without a person driving a browser?
You still set the rules
Selling to agents does not mean giving up control. Prices, stock reservations, discounts, and approval thresholds stay in your rules. Large orders or unusual terms can still go to a person before they are confirmed.
How to start
You do not need to rebuild your commerce stack.
- Pick one catalogue segment. A category with clean data and clear margins is enough.
- Expose search and detail as structured data. That is the minimum an agent needs to compare you.
- Add transactions behind rules. Cart, stock, price limits, human approval above a threshold.
- Publish a tool layer, such as an MCP server, over what you already have, so agents can find you.
From there you can add a selling agent on your own channels, buying agents on the procurement side, or both.
To see what this looks like on your current stack, see our agentic commerce service or book a call.
What goes on autopilot first?
Tell us how you work today. We will tell you where to start.
Book a call