Walmart's Agent Gateway: Why Retailers Are Splitting on AI Shopping

TL;DR: Walmart has launched an MCP-based system to let AI agents shop on its behalf, while Amazon blocks such agents, marking a pivotal split in how businesses handle autonomous commerce.

The retail industry is facing a fork in the road regarding AI agents. While Amazon has moved to block AI agents from completing purchases on its platform, Walmart is taking the opposite approach. The company recently announced the launch of Agent Gateway, a new infrastructure layer designed to let approved AI agents shop on behalf of customers. This isn't just a minor feature update; it is a strategic bet that controlled, permissioned access to AI agents will drive adoption rather than cause chaos. For business owners and tech professionals watching the AI space, this split offers a clear lesson on how to manage autonomous systems. It highlights that the future of agentic commerce depends less on raw capability and more on the trust framework surrounding it.

The Split: Open vs. Closed

Walmart CEO John Furner unveiled the Agent Gateway at the Masters of Scale Summit in San Francisco. The system supports agentic shopping at both Walmart and Sam's Club. The core technology is an MCP server, which allows agents to connect to Walmart's systems over the Model Context Protocol. This protocol, originally developed by Anthropic, has become the standard way for agents to interact with external tools and services. By adopting it, Walmart is signaling that it views MCP as the reliable rail for future transactions. In contrast, Amazon and various travel and booking sites are responding to the rise of consumer agents by blocking them entirely. The difference in approach is stark. Walmart is building a bridge for agents to cross, while Amazon is raising a wall. The outcome of this divergence will likely determine which model becomes the industry standard for the next decade of retail.

How Agent Profiles Work

Walmart is not just opening the door to agents; it is installing a security system. The company is rolling out Agent Profiles in November. This feature allows customers to authorize each agent with a dedicated profile. Think of it like adding an authorized user to a credit card. When you use a specific AI agent for the first time, a profile is created and linked to your Walmart account. You can delete it at any time. Within that profile, you can set spending limits, require approval for every transaction, or view agent purchases separately from your own. The system also restricts what the agent can see. Agents cannot access payment information or data beyond what is needed for the specific task. This granular control is designed to reduce the hesitation that shoppers feel about letting a machine handle their money. It maps an unfamiliar risk, autonomous purchasing, onto a familiar control model, making it easier for users to trust the process.

The Role of MCP in Agentic Commerce

For non-technical readers, the Model Context Protocol is the connective tissue of the modern AI agent. It is the standard language that allows an agent to understand how to use a tool, such as a shopping cart or a calendar. Walmart's adoption of MCP for its Agent Gateway is a significant validation of the protocol. It suggests that MCP is no longer just a developer tool for coding assistants; it is becoming the default infrastructure for business transactions. This matters because it means that the same protocol powering your coding agents is now powering your shopping agents. The infrastructure is converging. If you are building agents for your business, you are likely already using MCP to connect them to your internal tools. Walmart is simply applying that same logic to a massive public-facing platform. The stability of the protocol is what makes this possible. Without a standard way for agents to talk to services, the idea of agentic commerce would be fragmented and unreliable.

Trust and Verification at Scale

Walmart's approach highlights a critical issue in the AI agent ecosystem: trust. The company is introducing specific terms for AI agents that govern their behavior. Non-compliant agents will be blocked. The returns policy also applies even when an agent makes a mistake. This is a direct response to the fear that if an AI makes a bad purchase, the customer will be stuck with the loss. By guaranteeing returns and restricting agent access, Walmart is reducing the liability risk for the consumer. This mirrors the challenges faced in the developer agent skills market. When you install a skill for your coding agent, you face a similar trust issue. Can you trust that the skill will do what it says? Published research has found that roughly a quarter of agent skills carry at least one vulnerability. The lesson from Walmart is clear: capability without verification is risk. Whether you are letting an agent buy a TV or write code, you need to know what it can access and how it behaves. The control models being built in retail are the same ones needed in development.

Bottom line

The split between Walmart and Amazon is not just a retail dispute; it is a test case for how businesses will handle AI autonomy. Walmart's bet is that if you give users control, they will use agents. Amazon's bet is that the risks are too high to manage. For now, the data suggests that trust is the bottleneck, not the technology. If you are adopting AI agents in your own workflows, look at Walmart's model. Define the permissions. Set the limits. Verify the capabilities. The infrastructure is ready. The question is whether you are ready to let go of the reins, or if you prefer to keep them tightly in hand. Either way, the era of agentic commerce has arrived, and the rules of the game are being written now. If you want to explore how agent skills work in your own stack, you can browse the free directory at SkillzBench to see verified skills and their benchmarks.

Key takeaways

• Walmart's Agent Gateway uses the Model Context Protocol (MCP) to allow approved AI agents to search inventory and complete purchases.
• Amazon is currently blocking AI agents from making purchases, creating a stark contrast in retail strategies.
• Walmart introduces 'Agent Profiles' to let users set spending limits, require approval, and track agent purchases separately.
• The move validates MCP as a standard for connecting agents to services, mirroring the infrastructure used in developer agent skills.
• Trust and control are the primary barriers to adoption, with Walmart's safeguards designed to mimic familiar credit card authorization models.