Shopify opens checkout to AI agents: commerce starts being designed for machines, not humans

By Cadena 3 Argentina · September 28, 2026.

Shopify has expanded support for WebMCP —a protocol that lets a web page expose structured functions which an AI agent operating inside the browser can read and execute directly, instead of parsing HTML designed for human eyes— to cover the entire checkout process, not just product search. As the company detailed, any eligible merchant on the platform can now let a shopping agent review the order summary, change the shipping address or delivery option, and confirm the transaction, always with the buyer's explicit authorization.

The technical piece is three new tools: get_checkout, update_checkout and complete_checkout. Gil Greenberg, a product manager at Shopify, summed it up in a post on X: if an agent operates inside the buyer's browser, it should use these tools rather than “browsing” the page as a person would. Shopify already had an MCP server (Model Context Protocol, the standard for connecting an AI to external data and tools) so that agents could operate server to server; WebMCP is the missing piece for agents acting inside the user's own browser, built on the Universal Commerce Protocol (UCP), a common language for searching products, building carts and paying across different stores and agents.

The announcement does not come alone: the same day, Shopify confirmed a partnership with Instinct, one of the shopping agents already running on its infrastructure alongside Muse, according to the company.

As industry context, this fits into a broader race to become the default payment rail for agentic commerce: assistants that buy on the user's behalf. Over the past year, several payment networks and major AI platforms have unveiled their own components to let an agent authorize payments without direct human intervention at every step, and OpenAI itself has already tested an instant checkout built on Shopify infrastructure. The fight is less about having the friendliest assistant than about who processes —and bills for— the final transaction.

That is what matters about the move, beyond the announcement itself: Shopify is not competing to build the best shopping agent, it is competing to be indispensable to any agent, whoever owns it. It is the same logic we already saw in the struggle among big tech companies over agent standards: the advantage does not necessarily belong to the smartest model, but to whoever controls the “plumbing” through which the money and the data flow. Shopify, with its broad base of already-connected merchants, starts with a distribution advantage over the AI labs themselves.

In the short term, this raises uncomfortable questions that the original story does not resolve and that should not be downplayed. Who is liable if an agent buys the wrong size, updates a shipping address incorrectly, or falls for a fraud attempt disguised as a legitimate store? The buyer's authorization reduces the risk, but it shifts much of the oversight —today carried out by a person reading the checkout screen— onto software that interprets intentions. Conversion and user-experience teams, used to optimizing the purchase funnel for the human eye and thumb, will have to redesign that same screen so a machine reads it correctly: a change of craft, not necessarily of headcount, but a real one.

In the long run, however, the direction fits the underlying thesis we hold at Zendoric: the more administrative friction —comparing prices, filling in forms, chasing the best shipping option— people delegate to a trusted agent, the more free time is left for what really matters. Frictionless commerce is not a cosmetic luxury: it is one more piece of the abundance AI promises, provided the payment infrastructure is robust and auditable. The condition for that future to be benign rather than a new bottleneck is that this agentic payment rail does not end up in the hands of two or three companies with no real competition or oversight, something that, for now, neither Shopify nor its rivals have solved.