For decades, ecommerce was built on a simple premise: users visit, browse and decide. In 2026, that premise is starting to shift. Google has launched Universal Cart and expanded the Universal Commerce Protocol (UCP), an open infrastructure that enables AI agents to discover, compare and prepare purchases on behalf of users.

The question is no longer whether agentic commerce will happen. The question is who controls the transaction when a growing share of discovery, comparison and decision-making starts to be delegated to an agent.

From the web to protocols: when commerce starts speaking to machines

In January 2026, Google introduced UCP at NRF, developed together with Shopify, Target, Wayfair and Walmart, and backed by Visa, Mastercard, American Express, Stripe and Adyen. In May, at Google I/O, it announced Universal Cart: an intelligent, cross-surface shopping cart that works across Search and Gemini, with YouTube and Gmail expected to follow as future surfaces.

The standard is open source and designed so that any AI platform can adopt it. Its aim is not to create a new marketplace, but to provide a common language that allows agents, merchants and payment providers to communicate with one another. UCP structures key elements of the purchasing process —catalogue, availability, price, checkout, authentication, payment and post-purchase management— so that an agent can interact with them reliably.

Universal Cart is the most visible layer of that logic. It allows users to add products from different merchants, compare options, receive price or availability alerts and either complete the purchase with Google Pay or continue on the merchant’s own website. Google maintains that the brand remains the merchant of record, but discovery, comparison and purchase preparation increasingly begin to happen within platform-controlled environments.

Three maturities arriving at once: the paradox of perfect timing

Agentic commerce is not emerging from a single innovation. It is the result of three conditions converging at the same time. Language models have reached a level of reasoning that allows them to manage product comparisons across multiple criteria. Payment infrastructures —Google Pay, Stripe, Adyen— already support increasingly automated authentication and payment experiences. And users are beginning to incorporate artificial intelligence into different stages of their shopping journey.

The data reflects the scale of the movement, even if it is still starting from an early base. According to Adobe Analytics, traffic referred by generative AI tools to retail websites in the United States grew by 693% year-on-year during the 2025 holiday shopping season. Shopify, meanwhile, reports that in the first quarter of 2026 AI-driven traffic to stores on its platform increased eightfold, while orders coming from AI searches grew almost thirteenfold. Salesforce points in the same direction: 39% of consumers, and more than half of Gen Z, already use AI to discover products.

UCP does not invent agentic commerce. It gives it a common language, enabling the ecosystem to adopt it without every player having to build its own infrastructure from scratch.

When the point of sale becomes a protocol layer

With UCP and Universal Cart, the structural shift for retailers and brands is clear: the point of discovery is no longer only the website or the brand’s own app, but the agent’s surface.

Google preserves the merchant as the party responsible for the sale, but discovery, comparison and purchase preparation increasingly take place within its own environments. Brand loyalty, user experience and sales attribution need to be rebuilt in a context where the agent makes decisions based on structured catalogue data, price, availability and inventory.

For startups, this opens up a specific infrastructure opportunity: identity layers that connect users with loyalty benefits when they buy through an agent; fraud detection for autonomous transactions; multichannel attribution; catalogue data enrichment; and post-transaction loyalty models.

The infrastructure is ready; trust is not

Real adoption still faces significant friction. Strong growth in AI-driven traffic does not mean users are ready to fully delegate the purchase. Walmart detected conversion rates three times lower in purchases made through conversational interfaces than on its own website —a clear signal that the experience still does not solve every critical moment in the shopping journey.

Payment, identity and authentication infrastructure is not yet fully optimised for autonomous agents. Users may be willing to use AI to discover products, compare options or receive recommendations, but delegating the full execution of a purchase requires a higher level of trust. Conversion data shows a real gap between intention and execution.

Regulation around liability in autonomous transactions is also still undefined, adding uncertainty for both merchants and platforms. Who is responsible if the agent buys the wrong product? How can price, stock and conditions be verified? What happens in the event of a return, fraud or dispute?

Agentic commerce is not a new interface layered on top of the same model. It is a shift in the underlying layer: from traffic to catalogue intelligence, from store design to the quality of structured data, from user experience to delegated authorisation logic.

The strongest opportunities are not in replicating what Google is already building, but in solving what Google has not yet solved: portable identity, attribution in environments without third-party tracking, agentic fraud prevention and the reconstruction of the direct customer relationship after the first agent-mediated purchase.

Return to the GCO Ventures blog to keep reading about the investment trends reshaping digital commerce infrastructure and other analyses across our strategic verticals.