Agentic Commerce Comes of Age: What It Is and Why it Matters
For the better part of three decades, the governing metaphor of digital commerce has been the shopfront. A business builds a website, draws visitors to it through search and advertising, and does what it can to convert their attention into a purchase once they arrive. Almost everything in the modern commerce stack, from search engine optimisation through to the careful choreography of the checkout page, has been arranged around a single assumption: that a human being will, sooner or later, land on a page the business itself controls. That assumption has now begun to loosen, and the loosening deserves rather more attention than it has so far received in most boardrooms.
The reason is the arrival, in earnest, of agentic commerce: the practice of allowing AI agents to discover products, to evaluate them, and to complete a purchase on a buyer's behalf, frequently without that buyer ever visiting the seller's website at all. It is a development that has moved, over a remarkably short period, from conference demonstration to live consumer reality, and the businesses that understand it early will be considerably better placed than those who wait for it to become unavoidable.
What agentic commerce actually is
At its simplest, agentic commerce describes a transaction in which an AI assistant acts as the intermediary between buyer and seller. The buyer expresses an intention in plain language, perhaps asking for a navy wool overcoat under two hundred pounds that can arrive before the weekend, and the agent then performs the work that the buyer would once have undertaken in person: it searches, it compares, it selects, and, increasingly, it pays. For any of this to function, the seller must make its catalogue legible to the agent and its checkout callable by it, which is precisely the problem that the recent wave of infrastructure has set out to solve.
Beneath the surface sit open protocols, the Agentic Commerce Protocol introduced by Stripe and OpenAI in September 2025, and Google's Universal Commerce Protocol which followed soon afterwards, that give agents and businesses a shared language for discovery, checkout and payment. Layered on top of those protocols are commercial products, of which Stripe's Agentic Commerce Suite is the most prominent, that spare a business from having to build a bespoke integration for every agent it wishes to sell through, an undertaking that could otherwise consume as much as six months of engineering effort for each new agent supported.
What matters most in these arrangements, particularly for any firm anxious about losing its grip on the customer, is that the business remains the merchant of record. It retains control over which products are offered, at what price, and how orders are fulfilled, while the agent handles the discovery and the mechanics of the sale.
This is no longer hypothetical, and in the United States it is already live
The development that ought to concentrate minds is that none of this remains theoretical. Over the first half of 2026 the capability has moved from announcement and limited preview into general availability in the United States. American consumers can now complete purchases inside ChatGPT, inside Microsoft's Copilot, where buyers in the US are already able to purchase from retailers such as Urban Outfitters and Anthropologie without ever leaving the conversation, and, through Google's partnership with Stripe, inside AI Mode and the Gemini app. Meta, for its part, has enabled native checkout within advertisements on Facebook, so that discovery and purchase now occur in a single uninterrupted motion.
The behavioural evidence is moving in step with the infrastructure. Visa reports that close to half of American shoppers already use AI tools for at least one shopping task, and anticipates that millions will use agents to complete purchases over the coming holiday season. The shopfront, in short, is no longer the only place at which a sale can occur, and in the United States the alternative is not a forecast but a functioning channel. For businesses elsewhere, including here in the United Kingdom, the American market is best read as a preview of what is shortly to arrive, and as an invitation to prepare while there is still time to do so calmly.
Why it matters for B2C
For business-to-consumer brands, the implications are immediate, and for the unprepared they are uncomfortable. When a customer asks an assistant to recommend a product, the answer the assistant returns becomes the new shelf, and the brands that are surfaced enjoy something close to the visibility that a prime position in a physical store once conferred, while those that are absent are simply not in the running at all. Two things, broadly, determine whether a brand appears. The first is the quality and structure of its product data, since an agent can only recommend what it is able to read, which means that clean, current and well-described feeds, with accurate pricing and availability, have quietly become a commercial asset rather than a technical afterthought. The second is whether the brand features in the model's understanding of the category in the first place, a question of generative visibility that is important enough to warrant separate treatment, and which we address in a companion piece.
The risk that consumer brands should sit with is disintermediation. If the agent becomes the buyer's trusted adviser, then a brand that fails to earn a place in the agent's recommendations loses not merely the individual sale but, over time, the relationship itself, and relationships are a great deal harder to win back than transactions.
Why it matters for B2B
The business-to-business case is the less obvious of the two, and for that very reason the more interesting. Much of the early attention has settled on consumer retail, and yet the structural features of B2B commerce, the repeat purchasing, the standing catalogues, the negotiated pricing, the approval workflows and the broadly predictable reordering, make it unusually well suited to agentic handling. A procurement agent that can reorder consumables once stock falls below a defined threshold, assemble a compliant shortlist of suppliers against a written specification, or prepare and submit a request for quotation, removes a considerable quantity of the manual effort that currently surrounds routine purchasing, and does so in exactly the structured, rules-bound environment where agents perform at their best.
There is, moreover, a second strand of B2B agentic commerce that has little to do with traditional buying at all, namely the growing ability of agents and software services to pay one another directly for usage, through emerging machine-payment standards, so that an agent consuming an API or a data service settles the cost autonomously as it goes. For any business that sells to other businesses, whether goods, services, or access to software, the pertinent question is no longer whether machine-mediated purchasing will arrive, but how much of it the firm is actually positioned to capture when it does.
How enterprises should get ready
Preparing for all of this is less a single project than a sequence of deliberate moves, and they will be recognisable to anyone who has run a serious commerce operation. The foundation is product and catalogue data that is structured, accurate, current, and rich enough for an agent to represent the offering faithfully. Above that sits a decision about reach, namely which agent surfaces the business intends to support, a choice now made considerably easier by suites that connect to multiple agents and protocols through a single integration rather than demanding a fresh build for each. Then come the commercial and payment arrangements, including the firm's posture as merchant of record and the rails by which it will accept agentic payment.
Around the whole of it must sit governance, because selling through an autonomous intermediary raises a set of genuine questions that a prudent firm will want answered before, rather than after, it goes live: which products may be sold in this way, how pricing and promotions are to be controlled, how fulfilment will be handled, how fraud is to be detected, and how the agents themselves are authenticated and constrained. And beneath everything lies the prior question of discoverability, since none of this careful readiness counts for very much if the brand does not appear in the agent's answer in the first place.
The shopfront is about to stop being the only door
The channel is opening now, and it is opening fastest in the United States, where the infrastructure has reached general availability and consumers are already transacting through it. For businesses in the United Kingdom, the salient point is that the wait will not be a long one. Our expectation is that agentic commerce will arrive in the UK at scale over the course of the third quarter of 2026, which leaves organisations a window of only three to six months in which to stand up access to what is, in effect, an entirely new digital sales channel before the festive trading period, the very point in the year at which such a channel earns its keep.
The demand to justify that effort is already forming. According to the Agentic Commerce 2026 study reported by Checkout.com, nearly a quarter of UK consumers, some twenty-three per cent, already expect at least a tenth of their purchases to be AI-driven within the year, even as only three per cent of UK transactions presently involve an AI agent and fully eighty-nine per cent of merchants report that they are actively preparing for the shift. The prize at the end of it is considerable. By way of example, Juniper Research projects that global agentic commerce transaction value will climb from roughly eight billion dollars in 2026 to some 1.5 trillion dollars by 2030. A market expanding at that rate does not wait politely for the cautious to catch up.
The businesses that prosper from it are therefore unlikely to be those that move latest and most hesitantly, but nor will they be those that rush in without first preparing the unglamorous foundations of clean data, sound governance, and a genuine presence in the surfaces where buyers have begun to start their journeys. The shopfront is not about to disappear, but it is about to cease being the only door into the business, and the 2026 festive peak now bearing down on the calendar lends the question a deadline rather than merely a direction.
It would be wise, in short, to make certain the agents know how to find you. That, in turn, raises the question of discoverability we have deliberately set to one side, and it is the subject to which we will turn next. In a forthcoming piece we will look closely at generative engine optimisation: what it is, why it has come to matter a great deal, and what it can do for a brand's readiness, its discoverability, and its searchability in the eyes of the AI agents that are about to begin doing the buying.
Being found, after all, is the precondition for being bought.
