To Sell Through AI Agents, Merchants Need New Plumbing

U.S. retailers are preparing for a shift in how online transactions begin and execute. Agentic commerce is pushing purchase decisions into automated workflows that depend on accurate data and secure payments. “The battleground is shifting from persuasion to precision… You can no longer convince consumers to buy with just more ads or louder ads,” said Lou Keyes, CEO of Rithum at an event last month.

Stripe’s launch of its Agentic Commerce Suite shows how quickly this shift is occurring. The suite gives merchants a single integration that makes their products discoverable to AI agents, synchronizes price and availability, and handles checkout and payments inside agent environments without custom engineering.

In announcing the suite, Ahmed Gharib, product lead for agentic commerce at Stripe, said, “The Agentic Commerce Suite will help significantly expand the number of businesses that can start selling on AI agents.”

The foundation for these interactions is the Agentic Commerce Protocol, an open standard co-developed by Stripe and OpenAI that defines how an AI assistant can request product data, confirm availability, and submit a structured checkout request. It replaces improvised scraping or redirects with a predictable, verifiable format. Every inconsistency in that workflow creates risk for merchants: mismatched inventory, incorrect taxes, or ambiguous order provenance.

In presenting this shift, Adam Schwalb, Head of Product for Commerce at Stripe, said, “We’re building the infrastructure that lets businesses tap into the next major shift in how consumers shop.” That is, not creating new storefronts, but extending infrastructure so that agents can transact reliably across them.

A concrete example of these ideas in the wild came with Instacart’s Instant Checkout in ChatGPT, available to U.S. users. Shoppers could browse nearby grocery inventory and complete a purchase inside a chat interface, with Stripe powering the payment and ACP managing the structured exchange of order details.

Signals From the Frontline

Leadership across the ecosystem has started articulating how agentic commerce should work and what it should prioritize.

At Instacart, Chief Operating Officer Asha Sharma said, “Our goal is to make grocery shopping effortless, and integrating Instacart into ChatGPT brings us another step closer.” She makes the point that agentic commerce must solve real-world shopping problems rather than act as a novelty layer on top of them.

Nick Turley, VP and Head of ChatGPT at OpenAI, reinforced that dynamic by emphasizing seamlessness: “With the Instacart app directly in ChatGPT, users can go from meal planning to checkout in a single, seamless conversation.” His focus on reducing friction clarifies how agentic commerce differs from earlier attempts at conversational retail.

At Rithum, which integrates with Stripe to help brands expose optimized, AI-ready product data, Sean Meeks, President of Brands, said, “AI-driven commerce is accelerating fast, and the brands that show up early with the right data will win.” His emphasis on data fidelity aligns directly with the operational constraints that determine whether agentic commerce transactions succeed.

Where the Control Lies

As more commerce platforms adopt ACP, infrastructure that ensures fidelity, attribution, and trust becomes critical. Stripe’s earlier collaboration with OpenAI brought Instant Checkout for platform sellers into ChatGPT using a Shared Payment Token structure that protects buyer credentials and maintains merchant control over order flow. The mechanism limits exposure, scopes the token to the merchant, and provides a traceable payment event that merchants can audit.

This emphasis on merchant control has shaped new partnerships. Rithum’s integration with Stripe combines its catalog, availability, and order-orchestration systems with Stripe’s payment and checkout layer. Meeks’s earlier statement reflects Rithum’s belief that brands with ready, accurate data will benefit disproportionately from early adoption. The operational stack lets retailers present AI-ready data and accept transactions without building bespoke pathways for each agent surface.

At the data layer, provenance is emerging as a gating factor. AI agents rely on structured product feeds that remain accurate to the hour, particularly for price and availability. Rithum positions itself as the consistency layer that prevents product exposure inside AI systems from diverging from reality. Its role becomes critical because agent-initiated orders still flow back into the merchant’s systems.

What emerges from these developments is a picture that diverges from early rhetoric about autonomous shopping. The deciding factor is not whether AI agents can recommend products, but whether merchants can authenticate and fulfill agent-driven orders at scale with predictable risk profiles. Payments, tokenization, data provenance and attribution frameworks are central to the systems supporting these transactions. Stripe’s Agentic Commerce Suite and Rithum’s integration with secure checkout and catalog management illustrate how these components are being implemented.

In a previous announcement regarding ACP, Barbara O’Beirne, Global Head of Industries at Stripe, said, “AI is transforming commerce, and a new era of agentic commerce is beginning.” The collaboration with Rithum moves them another step forward.