Retailers have spent thirty years perfecting the art of selling to people online. Photography, merchandising, personalisation, loyalty mechanics, the psychology of the cart. It works because the shopper is human.
The next shopper, increasingly, is not.
Google's Gemini Spark, launched at I/O 2026, is an always-on agent that can be scoped to shop within user-defined budgets and merchant lists under the Agent Payments Protocol. Its early third-party integrations lean conspicuously retail. It is US-only for now, in beta, and requires human approval on transactions. None of that should comfort anyone. Every platform shift starts exactly this way, and every major AI company is shipping a version of the same thing. For retailers in Australia and everywhere else outside the launch market, the geographic delay is not a reprieve. It is a preparation window.
The bot judges different things
Here is the uncomfortable inversion: almost everything retailers currently invest in to win a sale is aimed at human perception, and almost everything an agent evaluates sits in the layer retailers habitually neglect.
Product data quality. An agent comparing five retailers for the same product needs complete, accurate, consistently structured attributes. Missing dimensions, inconsistent naming, sizes hidden in an image, stock status that lies. To a human these are irritations. To an agent they are disqualifiers, because it cannot evaluate what it cannot parse, and it will not guess when a competitor's listing is unambiguous.
Real availability and real pricing. Agents operating under spending mandates need firm numbers. Prices that only resolve at checkout, member-only pricing walls, shipping costs revealed at the last step. Retail grew comfortable with these frictions because humans tolerate them. An agent tasked with “buy the cheapest genuine option under $200 delivered” treats hidden costs as either an error or a deception. Neither wins the sale.
Checkout that a machine can complete. Every extra step, every CAPTCHA, every forced account creation is a place where an automated purchase dies. The industry spent years shaving seconds off human checkout. Machine checkout is a different problem with a different toolkit, including standards like AP2 that formalise how an agent, a merchant and a payment processor establish a verifiable transaction.
The reputation layer. Agents read your reviews. All of them, everywhere, weighted and cross-referenced. A retailer's aggregate public record becomes a standing input to every automated purchase decision.
The beautiful storefront problem
Having scanned retail websites across Australia, the UK and the US, the pattern I keep seeing is this: the correlation between how impressive a site looks and how well it performs under machine evaluation is weak, and sometimes inverted. Heavy, gorgeous, animation-rich storefronts often carry the worst technical legibility. Meanwhile some unglamorous mid-market retailers, running disciplined product data on boring platforms, are quietly the most agent-ready businesses in their category.
Size does not predict readiness either. Big brands assume their weight carries over. It does not, because the agent is not impressed by anyone.
What retail leaders should do this year
Not a replatforming. Not a moonshot. Three moves:
First, get a baseline. Score your digital presence the way a machine would: technical integrity, machine legibility, transactional accessibility, public sentiment. You cannot manage a channel you have never measured, and agent traffic is becoming a channel.
Second, fix the data before the funnel. Product data completeness and consistency is the highest-leverage, least sexy investment in retail right now. It improves your human experience too. Nobody has ever regretted it.
Third, assign ownership. Right now, machine readiness falls between ecommerce, IT and marketing, which means it belongs to nobody. The retailers that win the agentic shift will be the ones where someone's name is against it.
The shift from store to website took a decade and reshuffled the entire industry ranking. The shift from human shopper to agent-assisted shopper will be faster, because the infrastructure is already built and the incentive, for the customer, is convenience they can feel immediately.
The agents are learning to shop. The only question is what they find when they visit you.