The UK government has signalled that it wants clearer rules for payments made by AI agents, software that can research, choose and buy on a customer’s behalf. For a small online retailer, that hints at a shift in who, or what, arrives at your checkout, and it raises awkward questions about product data, payment authorisation and liability that are worth thinking through now rather than later.
Agentic commerce is the term for this: an AI assistant acting on standing instructions from a shopper, comparing options, then completing the purchase. It is early, and the timelines are long, but the direction of travel is clear enough to plan around.
What GoCardless is flagging
Open banking payments provider GoCardless has set out its reading of the government’s direction in its breakdown of an action plan for agentic commerce. According to that post, recent government and regulator publications point the same way: agentic payments are being treated as a priority for future policy work, and widespread agentic AI is discussed as a realistic prospect around the end of this decade rather than something imminent. Those are GoCardless’s characterisations of the documents it cites, so treat the detail as its summary rather than a direct quotation of any regulator.
GoCardless makes a specific prediction: consumers will hand recurring payments to AI agents faster than one-off purchases. Its reasoning is that anyone who has set up a Direct Debit has already delegated a repeating payment they do not actively supervise, so agentic commerce for subscriptions and bills is a smaller leap than for spontaneous one-off buys. GoCardless also reports that its own consumer research found appetite for AI managing recurring payments, provided shoppers keep control over limits and final approval.
The important point for merchants is the regulatory gap GoCardless highlights: the current payments rulebook assumes a human approves each transaction, and is not yet built for an agent executing payments on a pre-authorised, standing instruction.
What is actually changing
The framework itself is not being torn up. What is missing, in the language GoCardless uses, is the “plumbing” underneath: a way for an agent to prove whose behalf it acts on, to stay within agreed limits, and to be overridden the moment the customer wants to step back in.
GoCardless expects that gap to be addressed through future payments regulation work, covering three things in particular: legal liability when agents transact, “Know Your Agent” identity checks, and machine-to-machine authentication standards. Rather than rely on a single vendor summary, follow the primary work as it develops through HM Treasury on gov.uk and the Financial Conduct Authority.
Who is affected? Any business selling online, taking recurring payments, or listing products where price and availability change. If an agent cannot read your catalogue or trust your terms, it will simply choose a competitor it can.
What this means for your business
The blunt consequence is that a machine may become one of your customers, and machines are unforgiving about messy data. An AI agent comparing three suppliers will favour the one whose price, stock level and delivery terms are clean, current and machine-readable. Vague or out-of-date product information is not just a conversion problem any more; it is a reason to be skipped entirely.
Structured product data matters most. That means consistent titles, accurate prices including VAT, live stock status, clear delivery timeframes and unambiguous return terms an agent can parse without guessing. This is the same discipline that already helps human shoppers and search engines, so it is not wasted effort. If you have never audited how a machine reads your site, our explainer on the hidden blueprint behind your website is a sensible starting point.
Then there is the liability question, which GoCardless is right to leave open because the rules do not exist yet. If an agent buys the wrong size, the wrong quantity, or the wrong product entirely, who carries the cost of the return: the shopper, the agent’s provider, or you? Until clearer rules land, that risk sits inside your returns policy and your payment terms, so both need to be clear and defensible.
Payments are the sharp edge. Recurring rails such as Direct Debit already work on standing authority, which is why GoCardless expects them to adopt agentic behaviour first. If your model involves subscriptions or instalments, the transition may reach you sooner than if you sell one-off items.
What to do this quarter
- Clean up your product data. Make prices, stock and delivery times accurate and consistent across your store. If you are choosing or reviewing a platform, our comparison of Shopify, WooCommerce and Squarespace for UK stores covers how each handles structured data.
- Rewrite your returns and terms in plain, specific language. Spell out who bears the cost when the wrong item is bought, since agent-driven mistakes will land against whatever your policy says today.
- Understand your payment rails. Know whether your provider supports pre-authorised, limit-bound payments, and keep an eye on future payments consultations as they develop. Getting recurring billing right now pays off either way, as our guide to automating invoices and payments sets out.
- Tighten your monitoring. More automated buying means faster, higher-volume orders, so early detection of unusual activity matters. A simple online store security routine will help you catch problems before they compound.
None of this requires acting on hype. Agentic commerce is years, not months, from being mainstream, and the rules that will govern it are still being drafted. The work that prepares you for it, cleaner data, clearer terms and reliable payments, is worth doing regardless of when the agents actually arrive.
