On January 11, Google CEO Sundar Pichai stood before the National Retail Federation and announced the Universal Commerce Protocol—a new standard co-developed with Shopify, Etsy, Wayfair, Target, and Walmart to let AI agents buy products autonomously. Four days earlier, IBM released data showing 45% of consumers already use AI during their shopping journeys. Amazon's Rufus, meanwhile, now serves 250 million users and auto-purchases items when they hit target prices.
This isn't just another AI feature announcement. What's happening in Q1 2026 is the opening salvo in a war to determine who controls the interface between humans, their AI agents, and the entire commerce ecosystem.
Gartner predicts that by 2028, 90% of B2B buying will be AI-agent intermediated—pushing over $15 trillion through agent exchanges. The consumer opportunity is measured in additional trillions. But here's the uncomfortable truth buried in the hype: only 24% of consumers trust AI recommendations outright today, according to IBM's NRF study. The agents are ready to shop. The question is whether the infrastructure—payments, authentication, merchant data, trust, and human-defined guardrails—is ready for them. Platforms like AgentPMT are already building the guardrail infrastructure that makes autonomous agent commerce viable, with spending caps, vendor whitelisting, and real-time monitoring dashboards that give businesses control over what their agents can buy and from whom.
The Protocol Land Grab
The Universal Commerce Protocol (UCP) represents Google's bid to become the default interface for AI commerce. Co-developed with Shopify and endorsed by over 20 partners including Visa, Mastercard, Stripe, and Adyen, UCP defines product discovery, cart management, checkout, and post-purchase workflows in a single open standard.
The technical architecture matters. UCP uses layered extensibility: a core shopping service defines transaction primitives, capabilities add major functional areas (Checkout, Orders, Catalog), and extensions augment capabilities with domain-specific schemas. Merchants can implement their own bespoke functionality through the extension system. Agents discover what merchants support, negotiate what they can handle, and proceed to transact—all through structured APIs rather than scraping web pages.
But Google isn't alone. OpenAI embedded checkout directly into ChatGPT through its Agentic Commerce Protocol (ACP), built with Stripe and PayPal. The integration lets millions of ChatGPT users complete purchases using PayPal while connecting the platform to tens of millions of merchants.
Amazon, predictably, is playing defense. The company blocked outside agents from accessing its site—protecting its $56 billion advertising business—and launched Rufus as its own AI shopping interface. Rufus can now auto-buy items when they hit target prices for Prime members, with users saving an average of 20% through the feature.
The protocol wars of 2026 aren't about technical superiority—they're about distribution. OpenAI has 800 million weekly ChatGPT users. Amazon captures 40% of U.S. e-commerce. Google has deep retail relationships and search dominance. The winner won't be the best protocol; it'll be the one that's everywhere when agents start spending real money.
What Consumers Actually Want From AI Shopping
The IBM-NRF study released January 7 provides the clearest picture yet of how consumers actually use AI in commerce. Among the 18,000 respondents surveyed globally, 45% already turn to AI for help during their buying journeys—but the use cases reveal something important about trust.
Shoppers use AI to research products (41%), interpret reviews (33%), and hunt for deals (31%). Notice what's missing: autonomous purchasing. Only 24% trust AI recommendations outright. The remaining 76% verify before buying.
The gap between AI use and AI trust is the real story. Consumers want AI to shop for them—but they don't fully trust it yet. A Forrester survey found only about a third of consumers willing to complete payment through an AI answer engine at all, citing data privacy concerns. This is where human-in-the-loop architecture becomes essential—systems like AgentPMT's mobile app let users approve or reject agent purchases in real time, with push notifications that keep humans informed without slowing down the agent workflow.
Amazon's internal data tells a more optimistic story for the AI-trusting minority. Customers who engage with Rufus during their shopping journey are 60% more likely to complete a purchase. Monthly active users grew 140% year over year.
The companies that close the trust gap—through transparent recommendations, no hallucinations, honest pricing, and clear explanations of why an agent suggests what it does—will capture the market.
When Agents Become the Customer
The Gartner projection quantifies what's coming: by 2028, 90% of B2B purchases will be handled by AI agents, channeling more than $15 trillion in spending through automated exchanges. These systems will rely on verifiable data feeds and standardized trust frameworks that allow agents to negotiate, contract, and execute purchases at high frequency with minimal human intervention.
The competitive axis is shifting from "customer experience" to "agent experience." Brand storytelling and beautiful UIs matter less when agents optimize for price, availability, fulfillment speed, and return policies.
The problem? Most product catalogs were built for keyword search, not conversational AI. Agents need granular, standardized information—not just basic attributes, but detailed descriptions of fit, quality, durability, shipping timelines, and inventory status. They also need financial infrastructure—AgentPMT's per-tool pricing and per-workflow cost tracking let enterprises assign budget controls at the individual agent level, so each autonomous buyer operates within precisely defined financial boundaries.
Merchants that expose clean data become default suppliers. Those that rely on human emotional engagement will find themselves invisible to the growing segment of autonomous buyers.
Your product pages may look great to humans—but can agents parse your inventory data? Do you expose structured metadata? Is your pricing accessible via API? If agents can't find you or trust your data, they'll buy from competitors who made themselves machine-readable.
The Payment and Trust Infrastructure Gap
Mastercard and Visa are racing to build payment infrastructure that works for autonomous agents. Mastercard's Agent Pay requires agents to be registered and verified before payments, using "agentic tokens" that bind credentials to specific agents, devices, or merchants. Visa's Intelligent Commerce uses tokenization to limit credential exposure.
The authentication requirements are substantial: the agent must be identified, the consumer must authorize the agent, the payment provider must recognize both, and the merchant must approve the transaction. This four-party verification adds complexity but addresses a genuine security concern: AI shopping agents can be deceived by counterfeit merchants.
The fraud risks are real. Pindrop's data shows 3 in 10 retail fraud attempts are now AI-generated. Consumer security concerns match the risk: 83% of consumers in the IBM survey worried about security, even as 52% said they're willing to share information with AI assistants to get better recommendations.
This is exactly what we built AgentPMT to solve: a marketplace where agents can shop safely and securely within user-specified guardrails. Your agent discovers tools and services, transacts within budget limits you define, and never sees the credentials it's using. Credential isolation, blockchain audit trails on Base Network, and budget controls aren't nice-to-haves—they're prerequisites for participating in agent commerce at scale.
The Merchants Who Will Win (and Lose)
The competitive landscape is already separating merchants who optimize for agents from those who optimize only for humans.
Winners expose clean APIs, real-time inventory, transparent pricing, and predictable fulfillment. UCP's architecture makes this explicit: merchants publish profiles declaring what capabilities they support, and agents negotiate based on what both parties can handle. Merchants with incomplete or proprietary data simply don't appear in agent queries.
Amazon CEO Andy Jassy acknowledged the challenge directly: most AI shopping agents fail to provide a satisfactory customer experience. "They lack personalization and often provide inaccurate pricing and delivery estimates," he told analysts.
Microsoft's retail push underscores the infrastructure investments at stake. Adobe reported a 693% surge in AI-driven ecommerce traffic for the 2025 holiday season compared to 2024. Microsoft's Copilot Checkout lets merchants facilitate purchases directly within Copilot, supported by PayPal, Shopify, and Stripe.
This is why we built AgentPMT as a marketplace where agents shop—with DynamicMCP enabling cross-platform support so agents on Claude, ChatGPT, Gemini, or any MCP-compatible platform can access the same vendor catalog. Vendors list their tools and services once; agents discover them, use them, and pay per use. In an agent-mediated world, your visibility to AI systems is visibility to the market.
What This Means for Businesses, Builders, and Consumers
The era of AI-completed purchases has officially arrived—announced from the stage of the National Retail Federation by Google's CEO and validated by protocols from OpenAI, Shopify, Amazon, and every major payment processor. The $15 trillion B2B opportunity and multi-trillion consumer market isn't theoretical; it's being built now.
For builders: Your agents will need to transact. That means integrating with UCP, ACP, or both—or using a marketplace that already connects to the tools your agents need. It means building in budget controls, secure credential handling, and audit trails that payment providers will accept.
For vendors: Agents are becoming customers. If your products aren't accessible via structured data, if your APIs aren't agent-friendly, if you can't receive payment from autonomous systems, you're invisible to a growing segment of the market.
For everyone: The battle for the AI shopping interface is happening now. Google, OpenAI, Amazon, and Perplexity are each building their own vision of how agents interact with commerce. The protocols will likely converge—but the distribution won't.
What to Watch
UCP adoption velocity: How quickly do retailers beyond the launch partners implement UCP endpoints?
ChatGPT commerce expansion: OpenAI announced multi-item carts and more merchants coming. Track the Shopify integration rollout.
Amazon Rufus agentic features: The auto-buy feature is live for Prime members. Watch for expansion to more product categories.
Trust metrics: Will consumer trust in AI recommendations (currently 24% per IBM) increase as experiences improve?
Fraud incidents: Visa and Mastercard are building defenses, but the first major agentic commerce fraud will reshape the market.
The $15 trillion question isn't whether agents will shop—they already are. It's who controls the interface between agents and commerce, and whether the infrastructure exists to make those transactions trustworthy.
The agents are ready to shop. Are you ready to sell to them?
Explore the AgentPMT marketplace—where agents shop safely within user-defined guardrails.
Key Takeaways
- $15 trillion in B2B purchases will be AI-agent intermediated by 2028, according to Gartner—90% of all B2B buying
- Only 24% of consumers trust AI recommendations outright today, but 45% already use AI in their shopping journeys
- Protocol wars are about distribution, not technology: Google has retail relationships, OpenAI has 800M weekly users, Amazon has 40% of U.S. e-commerce