AI Agents aren’t waiting for retail to be ready

Why retailers must adapt to non-human shoppers

For the past 20 years, ecommerce has been built around one customer: the human.

Now a second customer of the internet is emerging. AI agents are beginning to search, compare and make purchasing decisions on consumers’ behalf, yet most ecommerce infrastructure simply wasn’t built with them in mind. Consumer research from Which? found that 51% of UK adults use AI search tools in their personal lives to search for products, services and advice online. And this is set to become increasingly autonomous, with Mastercard predicting that more than one in ten online shoppers will routinely use AI agents to shop and pay on their behalf by 2030.

Retailers are going through what I would describe as ‘corporate therapy’ right now. Businesses know agentic commerce is moving quickly, but there is no definitive playbook, so everyone wants to know what everyone else is doing. Where are they investing? What are they experimenting with? How quickly are they moving?

We can’t possibly expect to have all the answers yet. But retailers can start preparing for what it means to serve a customer that isn’t always human.

AI agents don’t window shop

An AI agent doesn't care whether your homepage looks beautiful, and it isn't going to be seduced by a clever banner or linger over a particularly nice piece of lifestyle photography. What it wants is information: is this the right product, is it in stock, what does it cost, and can it be delivered when the customer needs it? Most importantly, can it trust the information enough to recommend the product?

For example, a typical automative parts dealer has an incredible amount of fitment data that can establish whether a particular bulb, wiper blade, or component is right for the make, model, and age of a customer's vehicle. For an agent, that kind of clean, organised, and machine-readable information is enormously valuable because it provides enough confidence to say: this is the product you need.

But the experience also depends on what somebody is buying. If a customer is spending £3,000 on a Burberry trench coat, they are likely to want to take their time over the decision. Exploring rich imagery, considering the details and deciding for themselves whether it is worth £3,000, rather than handing that judgement over to an agent. If I need a Bosch wiper blade for a long drive this weekend, I don't need 15 beautiful pictures. I need to know it fits my car and will get the water off my windscreen.

Retailers should start thinking about their journeys in those terms. Which parts do customers genuinely enjoy and which are simply tasks they would happily hand over to an agent?

The heavy lifting starts with data

That makes some of the more mundane parts of retail technology much more important. Product specifications, inventory, pricing, fulfillment information and policies need to be accurate, structured and accessible. Yet for many established retailers, that information still sits across systems that don't necessarily communicate particularly well with each other.

Becoming agent-ready doesn't mean ripping out the entire technology stack tomorrow. The foundations are much more practical: clean up product data, organise catalogues, improve APIs and make sure stock, pricing and fulfilment information can be accessed and understood by the systems acting on customers' behalf.

Some surprisingly old-fashioned elements become useful here too. We have found that agents like detailed FAQs because they provide structured information and help establish trust. They also look to sources such as Reddit and Trustpilot for validation. What might feel like an overlooked corner of a website to a human can be valuable to an agent deciding whether a merchant or product should make its shortlist.

Being discovered is only one part of the challenge. If an agent selects a retailer only to find that the product isn't actually in stock or can't be delivered as promised, that could influence whether it chooses that merchant again. Being agent-ready means being discoverable, trustworthy, and able to deliver on what your data says.

Yesterday’s technology meets tomorrow’s shopper

There is no reliable timeline for how quickly agentic commerce will develop. It could be six months, nine months, 12 months, or 18 months before agent-to-agent transactions become a much more established part of shopping. That uncertainty doesn't make waiting risk-free.

Digital-native businesses can often experiment relatively quickly, while traditional retailers may be dealing with decades of technology investment, fragmented data and much larger, more complicated technology estates. But the work that prepares a retailer for agents – better data, connected systems and stronger infrastructure – has value regardless of exactly when the market reaches a tipping point.

It also gives retailers space to experiment while the market is still taking shape. Once significant transaction volumes start flowing through agents, everybody will want to respond. Doing the heavy lifting earlier means there is time to learn rather than scramble to catch up.

The ecommerce dashboard will need to change

Agentic commerce could also force retailers to rethink the metrics they have spent decades optimising. Customer acquisition cost, conversion, basket size and lifetime value make sense in a world where a human sees an advert, visits a website and moves through a recognisable funnel. When an agent can assess thousands of options, shortlist a product and potentially transact directly, that journey looks very different.

It creates new possibilities too. An agent could weigh up whether one retailer can deliver this afternoon, another offers a better price, and another provides useful membership benefits. Pricing, loyalty and fulfilment could all become dynamic factors in winning that transaction.

This is where effectiveness matters as much as efficiency. Efficiency is doing more with less. Effectiveness is using this technology to do what wasn't possible before: creating new opportunities, growing existing revenues, and potentially opening entirely new revenue streams. For retailers operating on thin margins, that is a much more interesting conversation.

Don’t optimise for machines and forget about people

In the rush to become machine-readable, retailers can't forget that human beings still quite enjoy being human. If agents take on more of the functional side of shopping – searching, comparing, checking availability and making routine purchases – genuinely human retail experiences could become more valuable, not less.

That means thinking about two experiences at once. For agents, the priorities are structure, availability, price, trust and utility. For people, brand, creativity, community and connection still matter enormously.

Gymshark's Soho location, for example, is developing a communal lifting space where people can train together, while AllSaints has talked about bringing live music back into its Spitalfields store. Those experiences aren't necessarily led by an immediate sale; they're about giving people a reason to spend time with a brand and with each other.

Brand doesn't disappear in an agentic world either. For agents, it becomes more about substance: why should it choose you, what value do you create for its customer, and what evidence is there that you can be trusted?

There is still plenty we don't know about agentic commerce, and retailers shouldn't pretend otherwise. But uncertainty isn't a reason to do nothing. Get the data in order, identify the customer problems agents can genuinely solve, experiment early, and start considering how success will be measured.

For two decades, we've built ecommerce almost entirely for humans. The next era will need to work for humans and agents at the same time, and the agents aren't going to wait around for retail to catch up.