Just in: Why Agentic Commerce Just Stopped Being a Payments Race and Became a Liability, Shelf and Rails Question

Who pays when an AI agent buys the wrong thing?

For a year, liability was the quiet blocker under every agentic commerce pitch. The technology could complete a purchase; the trust to let it could not keep up, with roughly four in five financial institutions expecting agent-driven fraud to rise.

That is why American Express becoming the first card network to underwrite agent error matters more than another protocol launch. Its Agentic Commerce Experiences developer kit pairs authentication and intent infrastructure with a purchase-protection pledge: if a registered agent buys the wrong item, size or merchant, an eligible cardholder is covered. Visa and Mastercard have been laying their own rails in parallel — Visa tokenising credentials through its OpenAI partnership, Mastercard scaling Agent Pay across more than thirty partners and into ASEAN.

The so-what for brands is simple. The moment a major network eats the cost of agent mistakes, the risk calculus for every merchant and consumer changes. Authentication and intent verification stop being a 2027 problem and become a payments conversation for this quarter. The brands sitting at the table as these rules are written will shape them; everyone else will inherit them.

Is your product even visible to the shopping agent?

OpenAI's first attempt at in-chat buying stalled quietly at fewer than thirty live merchants earlier this year — a useful reminder that AI-native checkout is far harder than a launch video suggests. The more telling move is what followed. Shopify has told its merchants their products will soon be discoverable and purchasable inside ChatGPT through agentic storefronts, shifting from a bolted-on checkout to native presence where the shopper is already asking. Google, meanwhile, has been widening agentic checkout in AI Mode since late last year, and on current form Perplexity and Google are the two most able to complete a sale inside an assistant.

Here is the shift brands keep underestimating. When an agent curates the journey, you stop winning the click and start needing to win the recommendation. Product selection moves from visual merchandising to API-driven logic. AI engine optimisation is becoming the new fight for digital shelf space, and being invisible to the agent is the new being invisible in search. The unglamorous truth is that this is a data-hygiene job — clean feeds, structured attributes, accurate availability and reviews — not a new campaign. Nominate an owner for it before the shelf hardens.

What happens when agents leave the marketplace they were born in?

Amazon rebranded Rufus as Alexa for Shopping in May and, for the first time, let it complete purchases beyond its own marketplace. Buy for Me locates a product on another retailer's site and completes the purchase using stored Amazon details; Auto Buy transacts when a shopper's target price is hit. By the close of the last holiday run the agentic shopping line was reportedly nearing twelve billion US dollars, with more than 350 million customers having used the assistant in the past year.

This is the walled garden reaching over its own wall. An agent buying from your site using a shopper's stored Amazon credentials quietly changes who owns the customer relationship, the data and the decision moment. For any brand with a direct channel, that is a distribution opportunity and a disintermediation risk at once. The most plausible near-term reality is that opting out simply makes you harder to buy, so the sharper question is how you protect first-party data and margin when the transaction is brokered somewhere you do not control. Model what a fifth of your demand arriving through an agent does to your economics, then decide your terms deliberately.

Why is Southeast Asia ahead — and what does it mean for Australia?

While Western markets argue card versus wallet, Singapore recorded the region's first authenticated agentic transaction in March, and Mastercard has been rolling AI-initiated payments into ASEAN. The structural reason is worth internalising: stored-credential wallets like GrabPay, ShopeePay and GCash are far easier to delegate to an agent than a raw card number, and real-time rails such as PayNow, DuitNow and QRIS already settle instantly and cheaply. Around forty-four per cent of Singaporeans say they would let an agent shop for them, and Asia-Pacific is on track to drive close to two-thirds of the world's new retail sales over the next five years.

For Australian leaders, the signal is adoption intent plus infrastructure investment. MI3's agentic AI report has boards weighing the same questions, the vast majority of local brand manufacturers and retailers expect agentic commerce to reshape their business, and Anthropic has opened a Sydney office courting enterprise and government on safety-first adoption. The takeaway for brands: resist a single global playbook. The likely path is that agent delegation matures faster in wallet-first Asia than in card-first Western markets, so sequence your rollout to the rails, not to headquarters.

The takeaway: treat this as an operating-model shift, not a marketing experiment

Pull these threads together and the pattern is clear. The agentic commerce contest has moved from building the agent to controlling the rails, the shelf and the liability around it. Retail media is feeling it too — a channel built on influencing shoppers before checkout looks exposed when an agent selects and buys, which is why measurement and clean-room capability are quietly becoming the battleground.

If there is one practical move to make this quarter, it is this: name a single owner across payments, data and commerce who is accountable for how your brand shows up to agents — how you are authenticated, how your products are read, and how you keep your customer relationship when the buying is brokered. The leaders who treat agentic commerce as a data, payments and operating-model question now, rather than a campaign later, are the ones who will still own their customer when the agent does the buying.

That is the shift worth watching. Not who built the smartest agent — but who built the business the agents cannot route around.