April 20, 2026

Benji Fisher's April infrastructure audit counted 4,014 stores with valid Universal Commerce Protocol manifests. Nine of them deliver a flawless end-to-end agent shopping experience. That is a 0.2% flawless rate, even as the verified store count grew 33% from roughly 3,000 in March. The B+ tier of 422 stores works most of the time. Agents can discover them, search their catalog, and usually reach checkout. But "usually" is not good enough when the agent is spending someone's money. Search works everywhere. Checkout is the bottleneck. Fisher's data shows 99.6% of stores support checkout declarations in their manifests, yet only 53% of agent-initiated checkout attempts actually complete on the best-performing model. Shopify accounts for 99.2% of the verified stores. BigCommerce, WooCommerce, and Magento together contribute seven. The ecosystem does not need more stores. It needs the stores it has to work more reliably.

Visa is betting that payment infrastructure will close part of that gap. On April 8, Visa launched Intelligent Commerce Connect, a network-agnostic and protocol-agnostic on-ramp to agentic commerce that enables agents to pay and merchants to accept agentic transactions through a single integration. The system supports Trusted Agent Protocol, Machine Payments Protocol, Agentic Commerce Protocol, and Universal Commerce Protocol through the Visa Acceptance Platform, meaning merchants do not need to integrate separately with each protocol. Visa Chief Product Officer Jack Forestell framed the stakes plainly: "I haven't seen anything like this since the dawn of ecommerce itself." Mastercard is moving on a parallel track. The American Banker reports that Mastercard executed its first agentic payment in Hong Kong via a ride-share booking, expanding its agentic commerce network to Hong Kong, Australia, the United States, and India. Both card networks have partnered with Stripe on Shared Payment Tokens and with Cloudflare's Web Bot Auth technology for agent identity verification. The same American Banker coverage surfaces numbers that quantify how fast consumer behavior is shifting. Forty percent of Americans have purchased items via AI agents that they would not normally have considered. Fifty-three percent of U.S. businesses would allow AI agents to negotiate prices with other AI agents, and 71% are willing to optimize their products specifically for AI agents.

Meta turned that willingness into a live product. Stripe's recap of Shoptalk 2026 reports that Meta launched a new Facebook checkout flow built on the Agentic Commerce Protocol. The flow takes a shopper from an ad click through product details and AI-generated review summaries to completed purchase without leaving the app. Stripe introduced its own Agentic Commerce Suite at the same event, enabling merchants to syndicate catalog and commerce infrastructure once across compatible agents rather than building separate integrations for each platform. OpenAI search data shared at Shoptalk revealed that more than half of searches on the platform are discovery-based, and 70% of those include constraints like price, size, or delivery window. That constraint density is what makes agent-mediated commerce structurally different from search-based shopping. An agent does not browse. It evaluates structured data against explicit parameters, and the merchants whose data satisfies those parameters are the ones that surface.

Delivery infrastructure is becoming one of those parameters. nShift's analysis of agentic commerce and fulfillment argues that delivery has stopped being a post-purchase function and become a pre-purchase ranking signal. When an AI agent compares two merchants offering identical products at identical prices, it selects based on delivery speed, cost, reliability, and pickup availability. Agents query delivery options programmatically through APIs, not by reading shipping pages. Retailers whose fulfillment data lives in static shipping tables, manual carrier assignments, and portal-only return processes are invisible to that evaluation. nShift frames the fix as normalizing tracking data into unified feeds, stress-testing delivery rules to protect margins, and auditing checkout for API readiness. The retailers that have already done this work are the ones agents recommend. The ones that have not are the ones agents skip, silently and without explanation.

Shopware shipped one of the clearest infrastructure responses this month. The Shopware 6.7.9.0 release introduced a native Agentic Commerce sales channel that lets merchants make their products available to platforms like ChatGPT through a compliant JSONL feed while tracking the business impact of AI-generated traffic. It supports multiple AI platforms from a single configuration and begins with an OpenAI product feed. Shopware built this as a top-level sales channel, not a plugin, which means agentic distribution sits alongside web and point-of-sale as a first-class commerce surface. Syndigo's Shoptalk 2026 analysis reinforces why this matters. Product data quality emerged as the universal dependency across every agentic commerce discussion at the conference. "High-quality product data remains the common currency that powers execution across all of it," Syndigo writes. The industry does not yet have universal ways for agents to transact cleanly across multiple merchants, but every path forward runs through structured, machine-readable product information.

The regulators have not kept pace with any of this. The Center for Data Innovation warns that regulations designed for human-operated systems are creating uncertainty that could slow agentic commerce adoption. Federal Regulation E, which governs electronic fund transfer disputes, lacks clarity on whether consumers retain dispute rights when an authorized agent violates instructions or purchases a mistakenly-priced item. Sarbanes-Oxley Section 302 requires executives to personally certify internal controls, but no guidance exists on whether AI operating parameters satisfy that requirement. Healthcare, food and beverage, and financial services operate under procurement rules that assume a human authorized each transaction. The Center for Data Innovation argues that the underlying principles of digital payments, authorization, consent, liability, and auditability, should largely apply. But "should apply" and "legally does apply" are not the same thing, and the gap between them is where enterprise adoption stalls. Visa and Mastercard can build the rails. Shopware can build the feeds. Fisher can count the stores. None of it scales until the legal frameworks catch up to the infrastructure that is already live.