Walmart will let customers connect their own AI agents to shop on its website starting next month, building an infrastructure layer the company calls Agent Gateway to manage agent transactions1,2. The move, detailed by Walmart President and CEO John Furner at the Masters of Scale Summit in San Francisco on October 21, 2026, positions the retailer opposite Amazon, which has blocked AI agents from making purchases on behalf of online shoppers.
Agent Gateway is a new MCP server built on Model Context Protocol, the standard originally created by Anthropic that lets AI agents connect to external data sources, tools, and workflows. The system will support agentic shopping across both Walmart and Sam's Club.
Controls baked into the protocol
Walmart is pairing openness with guardrails. By default, every AI-agent purchase will require customer approval. Customers can alternatively set a spending cap, such as up to $100, within which an agent may transact autonomously. Agents are barred from changing a customer's payment method and restricted to accessing only the information needed to complete a given task. Walmart said it will block agents that violate these terms.
A new feature called Agent Profiles will let users review agent-initiated purchases separately from their own orders and revoke agent access at any time.
Diverging retailer strategies
Amazon has taken the opposite approach, blocking Perplexity AI's shopping agent and Meta's Muse AI assistant from transacting on its platform. Various travel and booking sites have similarly locked out AI agents. Walmart, by contrast, had already partnered with Meta on Muse and offers integrations with ChatGPT, Gemini, and other AI applications.
The timing carries additional context. Meta suspended Muse's conversation-sharing feature on October 21 after a security firm found more than 6,800 user conversations exposed on the open web[1]. Separately, the AI-agent security market is splitting into a standalone category, with multiple vendors staking claims on different layers of the stack[3]. Trajectory-based evaluation methods are replacing static benchmarks as the trust gate for agents that execute sequential tool calls in the real world[2].
ANALYSIS By codifying agent behavior rules and requiring approval by default, Walmart is effectively proposing its own trust framework for agentic commerce rather than waiting for an industry standard to emerge. The retailer absorbs the integration cost but gains a channel that Amazon is refusing to serve.