February 9, 2026

Agents are flooding retail sites. Almost none of them buy anything. HUMAN Security, which monitors over twenty trillion digital interactions per week, tracked a 6,900% surge in agentic traffic through the first eight months of 2025, which HUMAN attributes to launches like ChatGPT Agents and Perplexity Comet. Of the pages these agents browse, 87% are product-related. But only 2.2% interact with carts and checkout flows. Discovery is solved. Purchasing is not. That gap defines the next trillion-dollar battleground, and every major platform, analyst firm, and payment network is scrambling to close it.

McKinsey's "automation curve" gives the clearest framework for understanding where we are. The progression runs from assist to assemble to supervised execution to full autonomy. Right now, agents research, compare, and evaluate. Humans still click "buy." The projections for what happens when that last barrier falls are staggering. McKinsey estimates $2.3 trillion in global retail sales transformed by agentic commerce this year. By 2030? $3 to $5 trillion. Morgan Stanley sees $190 to $385 billion flowing through US agentic channels by then. Bain projects 15 to 25% of all e-commerce agent-mediated. The forecasters disagree on magnitude. None of them disagree on direction.

The conversion advantage is not subtle. Users arriving via AI-generated answers convert 4.4 times better than those from traditional search, according to HUMAN citing Semrush. Adobe's own data shows 31% higher conversion from AI-optimized traffic. Even more dramatic, Salesforce customers using AI agents achieved 59% higher growth rates. The baseline is accelerating underneath these numbers. AI-driven ecommerce traffic surged 693% during the 2025 holiday season compared to the prior year. Morgan Stanley found that 23% of Americans bought something via AI in the last month. This is no longer a leading-indicator story. Retailers who cannot capture agent traffic through machine-compatible surfaces are already losing sales they will never see in their analytics.

A new category of agent is accelerating this shift. HUMAN calls them "agentic browsers," tools like Perplexity Comet that autonomously navigate merchant websites, searching, filtering, adding to cart, and completing checkout without human clicks. Unlike protocol-based commerce through UCP or A2A, agentic browsers consume the existing web storefront in a machine-first way. They don't need merchants to adopt new APIs. They just need simplified DOM structures, non-fragile checkout flows, and clear structured data. This creates an immediate opportunity. Retailers that optimize their existing sites for agent parsing can capture agentic traffic today, before deep protocol integration, simply by making their storefronts less hostile to machines.

The data readiness question has become existential. McKinsey's blunt warning, summarized by Jonah Santo, lands harder than any forecast. "If your catalog, policies, and value proposition are not machine-readable, agents, and by extension shoppers, simply will not find you." EKOM AI crystallized the tension every product team now faces. "Emotion sells to humans. Structure sells to machines. Both matter now." Retailers without UCP, AP2, A2A, or MCP integration risk becoming invisible to AI discovery engines by Q3 2026. That's not a distant threat. Seven months from now, if an agent can't parse your catalog, it will route your would-be customer to a competitor who made the investment.

The enterprise vendors smell blood. NRF 2026 was the moment SAP, Microsoft, and Workday all announced agentic AI products in rapid succession, each racing to become the default infrastructure layer. Gartner predicts 40% of enterprise applications will feature task-specific AI agents by the end of this year. Salesforce introduced Agentforce Commerce with two-way AI conversations via SMS, WhatsApp, and email, turning messaging channels into autonomous sales surfaces. Shopify merchants are reportedly auto-enrolled in Copilot Checkout, making conversational AI commerce opt-out rather than opt-in. Nobody asked the merchants. The best language model loses if it can't access your checkout flow. The race now is for the most deeply integrated agent fabric across the full retail stack.

B2B is where the margin is. Complex procurement, multiple stakeholders, and rule-heavy approval workflows make B2B transactions ideal for well-governed autonomous agents. Picture an agent that monitors inventory levels, detects a threshold crossing, generates a purchase order, queries multiple suppliers with RFQs, and compares not just price but lead times, SLAs, and ESG scores. HUMAN describes an even more radical pattern already emerging. Developers purchasing services directly from their code editor, with an agent inside VS Code configuring and buying cloud capacity based on CI/CD requirements and budget constraints. No browser. No procurement portal. Just an agent and a policy file. commercetools argues that B2B presents enormous agentic opportunities around order workflows, approvals, and contract negotiations. Legacy ERPs and compliance loads will slow adoption relative to consumer, but each deployment will move more dollars.

Gartner's long-range forecast is worth sitting with. By 2028, AI machine customers will replace 20% of interactions at human-readable storefronts. By 2035, 80% of internet traffic could be AI agents. Today's 2.2% checkout rate is a floor, not a ceiling. The retailers investing in machine-readable catalogs and protocol compliance right now are not preparing for some hypothetical future. They are competing for customers who have already arrived, and who will never open a browser.