February 2, 2026

eBay just flinched. The marketplace paused agentic shopping and blocked AI checkout activities entirely, citing governance and risk concerns around fully autonomous purchasing flows. The same week, Amazon went the opposite direction. Its new "Buy for Me" agent completes purchases from third-party sites when Amazon is out of stock, handling payment and shipping logistics on non-Amazon destinations while keeping the customer inside Amazon's experience. One incumbent retreated from autonomous transactions. The other turned itself into a meta-purchasing layer that treats the entire internet as its inventory. That divergence tells you more about the state of agentic commerce than any forecast.

Consider what "Buy for Me" actually does. A customer searches Amazon, finds an item unavailable, and an agent silently locates it on a competitor's site, completes checkout, and arranges delivery. Amazon has effectively absorbed the checkout funnels of smaller merchants into its own experience. The customer never leaves. The merchant? An anonymous fulfillment endpoint, invisible behind Amazon's interface. Retailers who spent years building brand loyalty through their own storefronts now face a scenario where their most common customer touchpoint is an Amazon agent placing orders on their site without any brand interaction at all.

Shopify is betting on the opposite architecture. The company released agentic infrastructure for cross-merchant cart building, allowing AI agents to compose carts spanning multiple Shopify merchants and optimize across price, shipping, and delivery times. This moves Shopify from powering isolated storefronts to acting as an agent-ready marketplace substrate. Amazon's "Buy for Me" centralizes the agent layer inside one company's ecosystem. Shopify distributes it across an open network of independent merchants who retain their brand identity. These are not two implementations of the same idea. They are competing visions of who owns the customer relationship when agents mediate every purchase.

New findings presented at the World Economic Forum in Davos add urgency to these positioning battles. Gartner now forecasts that AI agents will intermediate more than $15 trillion in B2B spending and approximately 90% of B2B purchases by 2028. The accompanying analysis was blunt about workforce impact, warning that "agents will change the nature of work, making some roles highly productive, and others obsolete" and urging organizations to begin skills transition programs this year rather than waiting for roles to evaporate in 2027 and 2028. Routine procurement tasks like RFP drafting, vendor comparison, purchase order generation, and shipment routing are expected to become fully agentic within that window.

The payment layer is being rebuilt from the ground up. Visa, Mastercard, and Stripe have all introduced agent-capable payment frameworks that authenticate AI agents as legitimate transaction initiators, encode spending constraints and merchant whitelists, and maintain auditable trails for disputes. The design shift is captured in a single line from Previsible. "The user becomes the approver, not the operator." Agents buy autonomously within constraints, then inform the human after the fact. This mirrors how B2B procurement policies have worked for decades with delegated purchasing authority, suggesting that consumer and enterprise agentic architectures are converging on the same trust model.

Microsoft wants to be the connective tissue. The company is rolling out agentic AI capabilities across the entire retail value chain, spanning merchandising, supply chain, store operations, customer service, and marketing. The shift is away from isolated copilots toward "coordinated systems that operate end-to-end" with fewer handoffs, faster decisions, and autonomous execution within guardrails. Picture what this looks like in practice. A demand forecasting agent detects a spike in a region. A supply chain agent reroutes in-transit shipments. A merchandising agent updates pricing and promotions. A store operations agent adjusts staffing. No human orchestrates the sequence. The agents coordinate among themselves, each acting within its domain, collectively functioning as a retail operating system.

The checkout funnel itself is disappearing. When agents handle the entire journey from intent to transaction, traditional metrics like cart abandonment lose meaning. Alakmalak projects scenarios for 2027 where a personal grocery agent negotiates prices with roughly twelve different stores and orchestrates automatic deliveries twice a week, ensuring a household never runs out of essentials. A predictive agent monitors a laptop's condition, tracks prices across retailers, and purchases a replacement at a 28% discount four months before the user would normally buy, scheduling delivery two weeks before the current device's predicted failure date. Shopping becomes invisible background infrastructure. The technical capability exists today. What doesn't exist are the data access agreements, trust frameworks, and liability structures to support it at scale. eBay looked at that gap and stopped. Amazon looked at it and jumped.

There is an underappreciated consequence to all this velocity. As logistics agents make routing decisions in near-real-time across jurisdictions, Vertex Inc. warns that the speed of these decisions may outpace human-managed compliance workflows. Compliance frameworks built around human decision cycles cannot monitor tens of thousands of micro-decisions per day. Tax automation, regulatory policy encoding, and audit trail infrastructure become prerequisites, not afterthoughts. eBay saw this problem and pulled the emergency brake. Amazon saw the same problem and decided to solve it in production. By this time next year, we will know which approach was right.