A new kind of visitor has started arriving at online stores, and it does not see the banner, the lifestyle photography or the carefully engineered urgency. AI shopping agents - software that takes a person's request, researches options, compares offers and increasingly completes the purchase - are moving out of demonstrations and into use, with dedicated protocols and agent-to-merchant networks forming around them. The share of commerce they carry today is small. But merchants have been here before: the arrival of price-comparison engines, of marketplace search, of mobile - each began as a curiosity, and each ended by rewriting what a storefront had to be good at. The agentic version of that rewrite has a specific character worth understanding early, because it inverts a great deal of what e-commerce currently optimises for.
Selling to an agent is a different discipline from selling to a human, in ways that are concrete rather than philosophical. An agent does not browse; it queries. It cares whether product data is structured, complete and machine-readable - accurate specifications, real availability, full pricing including delivery, explicit return terms - because that is what it compares. It is indifferent to persuasion and design, but unforgiving of ambiguity: the delivery cost revealed at checkout, the specification buried in a product photo caption, the stock status that turns out wrong are, to an agent, either invisible or disqualifying. And it changes the economics of trust: a human shopper absorbs a small disappointment and may return anyway; an agent that finds a merchant's data unreliable adjusts a ranking and consults it less, silently, at scale. In agentic commerce, data hygiene is merchandising.
The strategic stakes sit one level up, in a pattern this column has visited from another side: intermediation. Whoever operates the agent stands between the merchant and the customer, in the position search engines and marketplaces occupy today - deciding which offers are surfaced, taking a margin or a fee, and owning the customer relationship. The zero-click dynamic already visible in search, where answers replace visits, has a commerce sibling: the agent that purchases on a user's behalf may never show that user the merchant's site at all. That makes the coming years' negotiations - over agent access, over transaction protocols, over who verifies merchants and how disputes resolve - genuinely consequential for anyone selling online. Merchants collectively have an interest in open standards for agent commerce rather than a small number of proprietary gatekeepers, and little history of organising for that interest in time.
The practical preparation, fortunately, is mostly regret-free - the same work serves human customers, search visibility and agents alike. Get product data structured and truthful: complete specifications, real-time availability, all-in pricing, machine-readable markup of the kind already rewarded by search engines. Make policies explicit and findable - returns, warranties, delivery times - because agents weigh what they can parse and skip what they cannot. Watch your own traffic for automated purchasing and decide deliberately how to treat it, rather than letting bot defences make the decision by accident. And keep cultivating the one asset agents cannot easily reroute: customers who ask for you by name. A brand strong enough to be specified in the instruction - buy it from this shop - survives any intermediary. The store built entirely on being found by whoever happens past has the most to lose from a future in which the one doing the finding is software; the store whose data is clean and whose name is asked for has, once again, the least.