March 2, 2026

Walmart just put a number on it. Morgan Stanley reports that Sparky, Walmart's AI shopping assistant, drove a 25% increase in average shopper spend. Walmart deployed GenAI across the entire operation, with Sparky handling in-app and in-store product discovery while computer vision monitors shelves, LLMs manage inventory replenishment, and augmented reality tools supported holiday shopping. This is not a chatbot grafted onto a search bar. It is a multi-surface AI deployment with a measured, direct revenue impact. For an industry that has been stacking projections for months, that 25% figure is the first signal from a top-five retailer that AI shopping agents move money, not just traffic.

Agent surfaces are multiplying beyond apps and browsers. Google embedded shopping agents directly into Sea Group's Shopee platform, enabling customers to receive AI assistance from product discovery through checkout without leaving search results. The same coverage reports that Samsung's Galaxy S26 ships with Gemini and Perplexity alongside Bixby, turning the smartphone into an always-on agent host where users complete multi-step tasks like booking transportation with a single voice command. Consumers should "expect their next phone to complete complex tasks automatically in the background." Hardware OEMs are building agent hosting into device-level experiences that bypass traditional apps and browsers entirely. Every screen is becoming a potential purchase interface, and merchants optimized only for browser-based traffic are building for a shrinking share of how people buy.

The verification layer is getting specific. Checkout.com identifies "Know Your Agent" as the protocol that will determine which AI agents can transact and which get blocked. Analogous to Know Your Customer in banking, KYA requires merchants and payment networks to verify an agent's identity, permissions, and authorization scope before allowing a transaction to proceed. Network tokenization with agentic transaction identifiers lets processors distinguish agent-initiated purchases from human ones for fraud modeling and dispute resolution. The same Checkout.com analysis highlights Mastercard's Agent Pay, which embeds consumer-set spending rules and tokenized credentials directly into agent payment flows, and Visa's "AI-ready cards," which add FIDO security and spending controls designed for non-human purchasers. Payment networks announced agent-capable infrastructure earlier this year. Now they are shipping the verification layer to police it.

Enterprise procurement is producing the hardest ROI numbers in agentic commerce right now. Suplari reports its AI agents automate 60 to 80 percent of routine procurement work with accuracy above 90%, compared to under 80% from manual processes. The firm's documented results include 500% ROI with six-month paybacks, more than $3 million in annual value realization, and 75% faster contract cycles. Spend classification alone saw an 85% reduction in processing time. One Suplari customer discovered $150,000 in improved cash flow by having the agent analyze payment terms across thousands of vendor records and finding that nearly two-thirds of suppliers were configured for immediate payment rather than standard Net 60 terms. These are vendor-reported numbers from a company selling the product, and they should be weighted accordingly. But Suplari also reports that 90% of procurement leaders are implementing or planning AI agents within 12 months, which suggests the results are resonating beyond one vendor's case studies.

AI-to-AI negotiation is moving from concept to infrastructure. eComchain describes a framework where buyer and seller agents complete negotiations in seconds rather than days, with multiple buyer agents running in parallel on a single purchase. One optimizes cost. Another optimizes delivery time. A third evaluates supplier risk. As deadlines approach, agents adjust price thresholds within pre-set limits. Seller agents counter with dynamic pricing informed by inventory levels, demand signals, and competitive positioning. The architecture requires negotiation APIs, dynamic pricing engines, policy frameworks, structured protocols, and audit trails. Companies that can field seller-side agents will close deals at machine speed. Those that respond to RFQs manually will find the order has already been placed.

The workforce is already adapting. IBM is hiring three times more young employees than before, but their role is fundamentally different. They are "conductors" who manage what AI agents do, catch mistakes, and handle interactions requiring human judgment. The same AI Agent Store analysis reports that Entro Security launched monitoring tools specifically for autonomous agents that execute actions without human approval. About one-third of companies are testing agentic AI, with many expecting agents to handle most customer interactions within 18 months. But the analysis also warns that over 40% of institutional AI agent initiatives will be canceled by 2027. Governance and integration failures will kill more deployments than any technical limitation.

Walmart's 25% spend lift and Suplari's 500% procurement ROI are not projections. They are measured outcomes from deployed systems. The agentic commerce conversation is shifting from "will this work" to "who is already making money." That 40% failure rate draws the dividing line. Can an agent find your catalog, verify your terms, and close a purchase without a human fixing something along the way? Walmart can answer yes. Most retailers cannot.