In the past few months, I’ve seen a lot of hype about agentic commerce being ‘the next big thing for online retailers’, coupled with warnings to ‘get your site ready now or be left behind’.
While some are claiming that because of AI, e-commerce is undergoing its most radical transformation since the invention of the smartphone, I’m going to have to disagree.
As it stands, I think we are still a very long way off from agentic commerce becoming something people in the UK widely use to make purchases on their behalf, and even further from it being fully autonomous.
While the technology is there, there are a lot of barriers to adoption, so if your e-commerce business is flapping to get your website ready for agentic commerce (especially if that means shifting already stretched development resources into it), my advice for now would be to hold your horses.
In this post, I’ll explain what agentic commerce is and how you can prepare for it, but also outline the (many) barriers which will prevent mainstream adoption in the UK any time soon.
What is agentic AI?
While generative AI (i.e. a LLM like ChatGPT) simply answers questions, agentic AI systems are designed to perform tasks, not just output responses.
Think of it as the difference between a recipe book and a personal chef. A recipe book (generative AI) gives you information; a personal chef (agentic AI) takes your recipe, goes to the supermarket, buys the ingredients, and cooks the meal.
In an e-commerce context, an agent has agency: the power to navigate websites, use tools, and complete transactions independently to achieve a goal you have set.
How do agentic AI agents actually “shop”?
There are several levels to agentic commerce, from AI agents that browse the internet to consolidate products the user can review before buying them based on a given prompt without ever having to visit a website, to fully automated purchasing, usually for repeat orders, which require virtually zero human interaction, all the way through to AI agents negotiating with other AI agents for the entire journey, from discovery to fulfilment.
A recent report from McKinsey shows this automation curve, which defines these levels and their applications in e-commerce.
As well as understanding these levels and their different applications, to prepare your business for agentic commerce, you must also understand the internal logic of the agents now being deployed by tech companies like Google (Gemini) and OpenAI (Operator), as well as the protocols being used to make agentic commerce possible.
Unlike a search engine that returns results (both organic and paid links) for a user to click on and review one-by-one, an agent follows a recursive loop of reasoning and acting by reviewing and consolidating multiple sources.
If we take the level 3 defined by McKinsey as ‘Authorise’, the buying process begins with goal initialisation, where a user might say, “Find me a waterproof hiking jacket for under £200 suitable for a trip to the Peak District that fits in my backpack and can be delivered by Thursday.”
The agent then enters the discovery phase, querying product manifests and comparing technical specs, reviews, and delivery windows across multiple retailers.
In the execution phase, the agent doesn’t just show you a result; it calls a direct API or utilises “computer use” capabilities to physically navigate a website, select the correct size, and place the item in a basket. Finally, it uses a secure payment token to complete the transaction, logging the experience into its memory to refine future suggestions.
For this process to be fully automated, it requires a significant shift in user behaviour. The user must first establish delegated authority, which involves setting up an AI persona with specific brand exclusions and budget caps. In addition, the user needs to link a digital wallet to the AI and set strict “rules” (spending limits) so the agent can only buy what they’ve actually authorised. This ensures the AI doesn’t go nuts and spend loads of money or buy from a brand they don’t like when shopping on behalf of the user.
The protocols of the agentic web
For agentic commerce to work, the AI needs to be able to talk to the retailer and the payment provider in a language they all understand. Several key protocols are currently being developed to standardise this:
- Universal Commerce Protocol (UCP): Co-developed by Google and global brands including Shopify, Etsy, and Wayfair, UCP is an open-source standard designed for the full shopping lifecycle, from product discovery to final delivery. It allows merchants to publish their “checkout logic” so any AI agent can understand how to buy from them.
- Agentic Commerce Protocol (ACP): Launched by OpenAI and Stripe, ACP focuses on agent-native transactions, specifically enabling “Instant Checkout” within interfaces like ChatGPT. It uses a Shared Payment Token (SPT) to pass payment credentials securely without exposing the user’s actual card details to the AI.
- Model Context Protocol (MCP): Introduced by Anthropic (the company behind the LLM Claude), MCP acts as the open standard for AI agents. It allows agents to plug into a retailer’s live inventory, CRM, or shipping database safely, ensuring the AI is making decisions based on real-time data rather than outdated training sets.
- Agent Payments Protocol (AP2): This protocol focuses on the cryptographic “mandate” needed for secure payments. It provides the proof of consent that UK and EU regulators demand, ensuring that every penny spent by an agent is backed by a verifiable user instruction.
How to prepare your e-commerce site for agentic commerce
For a UK business to be “agent-ready,” there are several steps you need to take. Before deciding whether to proceed, carefully consider the impact it will have on your business beyond developing a roadmap for implementation.
Ensure your products can be found
You should prioritise product Schema.org markup using JSON-LD to provide deep context on price, availability, and return policies. Standardised identifiers like GTINs and SKUs are non-negotiable; agents rely on these to cross-reference data. Without them, your product cannot be included in the agent’s shortlist.
This is already best practice for on-page SEO (alongside having well-optimised product feeds for paid search), so make sure you have this in place regardless.
Simplify design and functionality
The next stage is keeping things simple in terms of design and functionality, which might mean disabling features that have worked well in the past, as well as limiting testing on the site (this means a lot of UX/CRO experiments will need to be halted).
Agents require “stable selectors.” If your “Add to Basket” button ID changes because you’re running an A/B test, the agent will fail and abandon the sale. And while pop-ups and gamification features might be a great way to boost sales, agents struggle with “Spin to Win” wheels, as well as complex CAPTCHAs, so providing a streamlined path for verified agents is essential.
Create a path to purchase just for AI agents
One way to get around this is to build infrastructure just for AI agents. This is known as a bot-friendly fast-track lane.
A fast-track lane is essentially a parallel version of your checkout flow designed specifically for authorised AI agents. Instead of forcing the agent to navigate a visual interface intended for humans, you provide a streamlined, data-only path.
How it works in practice:
- Bypassing interstitials: When a verified AI agent (identified via its unique User-Agent string or a secure token) hits your site, your system bypasses the “sign up for a 10% discount” pop-up and the “customers also bought” recommendations.
- Headless checkout: The agent is directed to a simplified checkout page, or even a direct API endpoint, where it can submit shipping details and payment tokens in milliseconds, rather than navigating through multiple steps or pages.
- Verified bot status: Using standards like the Model Context Protocol (MCP), your site can distinguish between “good” agents (authorised to buy) and “bad” scrapers (stealing data), allowing the good ones to move through the purchase process without being blocked by anti-bot security measures.
Decide which protocols are right for your business
Once the basics are in place, you’ll need to choose a protocol, and Google’s Universal Commerce Protocol (UCP) seems to be the best choice at the moment, though choosing the right agentic commerce protocol depends largely on which AI ecosystem (OpenAI vs. Google) your customers frequent and the level of autonomy you want to grant AI agents.
UCP is designed to work across verticals and is compatible with industry protocols such as Agent2Agent (A2A), Agent Payments Protocol (AP2) and Model Context Protocol (MCP). You can read the documentation and check out the ‘join the waitlist’ form here to get an idea of requirements and eligibility. If your website uses a platform like Shopify or WooCommerce, you can prepare for UCP where it is supported natively or via plugins.
If Stripe is your payment processor, you can prepare for the Agentic Commerce Protocol (ACP) by checking your platform settings for agentic features and applying for OpenAI’s “Instant Checkout.”
Agentic commerce in action
The US is currently the testing ground for agentic commerce. At the National Retail Federation’s annual conference in New York, Google announced its Universal Commerce Protocol (UCP) for agentic commerce in partnership with Walmart, Shopify, Etsy, Wayfair, and Target.
Just a month later, users were able to buy from retailers like Wayfair and Etsy directly within Google. This integrated checkout experience is available in the US via Google Search’s AI Mode and the Gemini AI agent.
– Screenshots showing UCP within Google AI Mode
It’s worth noting that at this stage, agentic commerce is not fully autonomous. The user still needs to choose the product and pay for it.
Barriers to the adoption of agentic AI
For now, fully autonomous agentic commerce is not readily available. This is for a number of reasons – most of which relate to the regulations that underpin the payments ecosystem.
Regulatory requirements
While the technology is ready, the legal landscape in the UK and EU is significantly more complex (and cautious) than in the US.
The EU AI Act, which comes into full effect for most businesses in August 2026, mandates strict transparency obligations. You cannot “trick” a user into thinking they are interacting with a human, and any AI-driven purchase must be “explainable” under consumer protection laws.
In June 2025, the UK passed the Data (Use and Access) Act (DUAA), which came into effect in February 2026. This act makes it easier for AI agents to access data, but it also places a heavy burden on businesses regarding transparency.
The most significant hurdle remains Strong Customer Authentication (SCA). Under the UK’s Payment Services Regulations (PSR), a transaction usually requires “active consent” at the point of sale.
For an agent to shop while you sleep, you need a pre-authorised mandate through open banking-enabled Commercial Variable Recurring Payments (cVRP) – and this isn’t even in place yet. Without this legal framework, an agent will simply get stuck at the 3D Secure (3DS) screen every time.
Trust in AI
Beyond the law, there are substantial trust and setup issues to overcome. Recent research from the Retail Technology Show (RTS) highlighted:
- The trust threshold: While adoption of AI assistants has doubled, 60% of UK shoppers remain mistrustful of using AI agents for end-to-end shopping missions.
- The control demand: Despite being perceived as one of the most tech-savvy demographics, roughly 70% of Gen Z shoppers insist on retaining “final control” over payment authorisations rather than granting full autonomy.
- Onboarding fatigue: Setting up an AI agent, which requires linking wallets and defining permissions, is time-consuming. For many, the “setup cost” currently outweighs the perceived convenience.
The liability loophole: who pays when the AI goes rogue?
If an agent “hallucinates” a purchase or buys an item based on incorrect data, the question of liability is complex. Under the Consumer Rights Act 2015, you, as the merchant, are generally liable if your site’s metadata was misleading.
However, the revised EU Product Liability Directive now treats AI as a “product,” meaning if the agent’s code is defective, for example, ignoring a “do not buy” command, the developer of the AI could be held strictly liable.
For the consumer, the risks are equally high. If you grant an agent “full autonomy” without setting budget caps, UK banks may argue you were grossly negligent, which could waive your right to a chargeback (where you dispute a charge on your card or bank statement and request a refund directly from the issuing bank) under Section 75 of the Consumer Credit Act. Once agentic AI does enter the mainstream, where the “liability gap” is likely to be tested in UK courts, making robust audit logs and clear permission sets becomes a business necessity.
What comes next?
Agentic commerce is coming – there’s no doubt about that. And with that in mind, there’s no harm in keeping up to date with the latest developments.
There may be faster adoption of agentic commerce in B2B, wholesale, and subscription-based businesses, where it is easier to define clear rules for when an AI agent can execute a transaction without permission from a human.
But just how much of a game-changer it will truly be is up for debate. I’ll be keeping a close eye on how things develop over the pond before I start making any recommendations for e-commerce clients to invest time and resources into preparing for it.
For UK B2C retailers, the opportunities that agentic commerce may bring sound exciting, but the barriers to widespread adoption appear to be too high for now.