Updated
What is agentic commerce, and is your website ready for it?
The 30-second answer
Agentic commerce is buying and selling where an AI agent acts for the customer. The customer states the intent in natural language, the agent does the research, compares items across multiple retailer websites, and can complete the purchase and payment within spending limits the customer controls. Stripe and OpenAI have already shipped an open standard for the payment half (the Agentic Commerce Protocol), and browsers are trialling WebMCP for the on-site half. For your business the question is practical: when an assistant shops on a customer's behalf, can it use your website, or does it buy from the company whose site it can use?
The numbers behind the noise
Two figures worth separating, both from McKinsey's October 2025 research on agentic commerce. The projection: by 2030 the US B2C retail market alone could see up to $1 trillion in revenue orchestrated by agents, with global projections of $3 trillion to $5 trillion. Treat that as a forecast, not a fact. The measurement: 44 percent of users who have tried AI-powered search already call it their primary and preferred way to search, against 31 percent who prefer traditional search. That second number is behaviour that has already moved.
The pattern for a UK company is familiar. Digital commerce has been through this before with search, mobile and marketplaces: the discovery habit changes first, the transaction follows, and the organisations that prepared while it was early kept the revenue. AI assistants are already answering buying questions today; the checkout side is what the protocols are being built for now.
The agentic commerce stack, in plain terms
Agentic commerce is not one technology. It is an ecosystem of three layers, and a business needs all three working together.
The risk of reading vendor content on this topic is that each vendor describes only their own layer as the whole story. A payments company will tell you agentic commerce is a checkout problem. A commerce platform will tell you it is a product-data problem. For most UK businesses the first real gap is simpler: the agent arrives at the website and cannot finish the task.
What this means for a UK business
The opportunity is conversion you currently lose. An agent that can check availability, get accurate pricing information and complete a task on your website converts on the first visit. One that cannot will either give up or buy from a competitor whose site it can use. Agent traffic is small today; the point of preparing early is that the work also fixes conversion problems human customers hit now.
The risk is losing control of the sale. When an assistant compares your product against three rivals and negotiates the choice, your brand strategy matters less than your data quality. Product information, availability and pricing need to be accurate, structured and accessible, because the agent reads them literally. Vague copy that a human shrugs past becomes a wrong answer in an agent's comparison.
Accountability stays with you. Sensitive actions need confirmation steps. Google's WebMCP guidance recommends confirmation dialogs on purchases, and the payment protocols are built around customer-set spending limits and explicit authorisation. Designing which actions an agent may take, and which need a human, is part of the implementation work, not an afterthought.
How to prepare, in order
A practical process for a UK company, smallest useful step first.
- Measure discovery. Run your buying queries through the AI engines and see whether your company is named. Free start: the AI visibility tool.
- Test the website with a real agent. Run an AI agent against your money journeys (quote, booking, purchase) and record where it stalls. This is the agent readiness audit, and it produces value even if you build nothing afterwards, because agent stalls are usually human friction too.
- Fix the data before the tools. Accurate product information, structured availability, visible pricing. For example, an agent comparing fire doors needs width, rating and price as data, not prose. An agent cannot be more accurate than the information you give it access to.
- Expose the actions that earn money as tools. 5 to 15 WebMCP tools with clear names, descriptions and schemas, confirmation on anything sensitive, tested against the Chrome and ChatGPT implementations.
- Watch the payment layer, adopt when your platform does. ACP support will arrive through the checkout providers. For most businesses this is a switch to flip later, not a build to start now.
Agentic commerce questions.
What is agentic commerce?
Agentic commerce is buying and selling where an AI agent acts for the customer: it researches options, compares products and prices across websites, and can complete the purchase and payment within spending limits the customer sets. The customer states the intent in natural language and confirms at the end.
What is the Agentic Commerce Protocol (ACP)?
ACP is an open standard developed by Stripe and OpenAI that defines how AI agents and businesses transact, including coordinating checkout and securely sharing payment credentials. It handles the payment side of agentic commerce. WebMCP handles the on-site side: exposing a website's actions as tools an agent can call.
How big is agentic commerce expected to get?
McKinsey projects that by 2030 the US B2C retail market alone could see up to $1 trillion in revenue orchestrated by agents, with global projections of $3 trillion to $5 trillion. Those are projections, not measurements. What is measured now: McKinsey found 44 percent of users who have tried AI-powered search already call it their primary and preferred way to search.
How does a UK business prepare for agentic commerce?
Three steps in order. First, be findable and cited by AI assistants, which is SEO and GEO work. Second, make your website usable by agents: clear product information, structured availability and pricing, and a WebMCP tool layer for the actions that earn money. Third, decide your controls: which actions need human confirmation, and what information an agent can access.
Is agentic commerce the same as WebMCP?
No. Agentic commerce is the whole category: agents choosing and buying for customers. WebMCP is one enabling technology inside it, a proposed web standard that lets a website expose its actions as tools. Payment protocols such as ACP cover the transaction itself. A business preparing for agentic commerce needs the discovery, on-site and payment layers to work together.
Sources
- McKinsey: The agentic commerce opportunity - published 17 Oct 2025, read 31 Aug 2026. Source of the $1 trillion US and $3 to $5 trillion global 2030 projections and the 44 percent AI search preference figure.
- Agentic Commerce Protocol - read 31 Aug 2026. The Stripe and OpenAI open standard for agent transactions.
- Stripe: What is agentic commerce? - read 31 Aug 2026.
- Chrome for Developers: WebMCP - published 18 May 2026, read 31 Aug 2026. The on-site tool standard and its confirmation-dialog guidance.
- Adyen: Agentic commerce and control - published 18 Feb 2026, read 31 Aug 2026.
Find out what an agent can actually do on your site.
The agent readiness audit runs real AI agents against your money journeys and hands you the stall points, for a fixed 450 pounds.