Just in: The Agentic Commerce Race Has Moved From Checkout to Control — What It Means for Brands Selling Through AI Agents

Something shifted this month, and it is easy to miss if you are still asking whether AI agents can actually buy things. That question is largely settled. The one that matters now is who controls the rails those agents shop on, who carries the liability when an agent gets it wrong, and whether the business on the other side can fulfil what an agent just ordered. For anyone selling through AI agents, that is a very different set of problems, and it is where the next eighteen months will be won or lost.

I have been watching the protocol jockeying, the enterprise data and the Australian policy conversation converge on the same point: agentic commerce has graduated from a demo into infrastructure. Here is what happened, why it matters, and the one move I would make this quarter.

Who owns the rails when an AI agent does the shopping?

The centre of gravity has moved from can an agent shop to whose plumbing it shops through. Google, with Shopify, is pushing its Universal Commerce Protocol as an open standard covering discovery through post-purchase, and it is bringing checkout into AI Mode and Gemini, per Google's own Marketing Live announcements and eMarketer's reporting. Opposite it sits the Agentic Commerce Protocol from OpenAI and Stripe, which powers discovery and merchant redirect inside ChatGPT.

The most instructive move of the year was OpenAI quietly retiring its in-chat Instant Checkout in March, roughly six months after launch, with barely thirty merchants live and the experience undermined by inaccurate pricing and inventory, as Digital Commerce 360 reported. It has repositioned around discovery and handed the transaction back to merchants.

So what for brands: the hard part was never the chat interface. It is clean product data and reliable inventory. Two camps are forming — one treats the AI surface as the shop floor, the other as the shop window with the register kept on the merchant's own site. Your feed quality, not your chatbot, decides which one works for you.

Are AI agents actually delivering inside enterprises, or just in demos?

Salesforce's second Agentic Enterprise Index, built from aggregated Agentforce usage between February 2025 and April 2026, is the clearest read yet. The average Salesforce business went from five activated agents to thirteen, deployments more than doubled year on year, build time more than halved, and businesses running agents saw four times higher retail online sales growth. Weekly employee usage tripled while escalation rates held steady — the trust signal that actually counts.

Set that against Gartner's standing warning that more than forty per cent of agentic AI projects will be scrapped by end of 2027 on cost, unclear value or weak controls, and its view that only around one hundred and thirty of the thousands of self-described agentic vendors are real.

So what for brands: both are true at once. Adoption is compounding and most projects will still fail, and the gap between them is governance and data readiness, not model quality. The leaders are treating agents as a measured workforce with owners, ROI and escalation paths — not as a science project.

What does Australia's National AI Plan mean for marketers?

The domestic conversation shifted from what agentic AI might do to what national policy now expects. Australia's National AI Plan has been read across the industry as a prompt to get serious, with Mumbrella calling it advertising's wake-up call. Underneath sits a sharper worry from the retail media side: Zitcha's Alberto Vergara describes a near future where agents scan, compare and buy for shoppers, upending how retail media networks sell inventory.

The blunt version is that if an agent makes the purchase decision, performance marketing as we have practised it starts to erode, because there is no human to serve the mid-funnel ad to. That lands on top of an Australian retail media market worth around two billion dollars and already wrestling with a measurement gap, as the IAB summit heard.

So what for brands: when the shopper is a model, brand salience and structured, machine-readable product data do the persuading long before an ad does. The most plausible outcome is that discoverability by machines becomes as important as ranking for humans.

Who vouches for the bot — and who is liable when it gets it wrong?

Two brakes on agentic commerce are being released at once. On the legal side, a US appeals court in early August overturned an injunction blocking Perplexity's shopping agent from Amazon, reasoning that because the agent acts only on a user's instruction, it is the user, not Perplexity, accessing Amazon's servers, per PYMNTS. It is the first federal appeals decision on agent access, and it tilts the ground toward agents being allowed in.

On the payments side, the card networks are answering the trust question. Mastercard's Agent Pay binds a tokenised credential to a specific agent, merchant scope and consent policy, and its June machine-to-machine extension launched with more than thirty partners; Visa's Trusted Agent Protocol uses cryptographic signatures to separate genuine agents from malicious bots.

So what for brands: the same logic that frees agents to shop pushes liability toward whoever set the agent loose, and roughly four in five financial institutions expect fraud to rise. Blocking agents by legal threat is a weaker play than it looked; authenticating and pricing for them is the stronger one.

What should leaders actually do about agentic commerce right now?

Pull the threads together and the picture is consistent. Bain sizes the US agentic commerce market at three hundred to five hundred billion dollars by 2030; Gartner expects one in five digital commerce transactions to run through AI platforms or agents by then; Adobe measured a jump of around four thousand seven hundred per cent in generative-AI referral traffic to US retail sites year on year. This is not a horizon-scanning exercise anymore.

If I had to pick one move, it would be this: treat your product feed as a board-level asset. Audit price and stock accuracy across every AI channel this quarter, because inaccurate data is the specific thing that sank a well-funded checkout. Then decide your posture on agent traffic deliberately rather than by default — authenticate the agents you want, detect the ones you do not, and structure your catalogue to syndicate cleanly across both major protocols rather than betting early on one.

The brands that win the agentic era will not be the ones with the cleverest AI. They will be the ones an agent can find, trust and reliably buy from — and then actually deliver to.