Walmart Just Built the Merchant Control Plane for AI Agent Shopping: Spending Limits, Agent Profiles and No Card Data

TL;DR

• Walmart’s Agent Profiles give AI agents separate identities inside a customer’s account, with agent purchases tracked apart from personal shopping.

• Every agent purchase requires customer approval by default. Customers can instead set per-agent spending limits, while Walmart may still request additional approval.

• Agents cannot access or change payment card information. Walmart’s Agent Gateway uses the open Model Context Protocol (MCP) to connect external agents to its shopping capabilities.

• The strategic signal for merchants and payment providers is clear: agent commerce needs its own controls for identity, spending authority and data access.

Walmart is building more than a new way to shop. With its announced Agent Profiles, Agent Gateway and agent terms, it is establishing a merchant-side control plane for agentic commerce: a system for recognising an agent, defining what it may do and keeping sensitive payment data out of its reach.

That distinction matters to fintechs, payment facilitators (PayFacs), POS providers, issuers and other payment businesses. The emerging question is not only how an AI agent initiates a purchase, but how merchants let it act on a customer’s behalf without treating it like a human with unrestricted account access.

Walmart announced the measures on 21 October 2026. The company’s announcement says Agent Profiles are expected to launch by late November.

What has Walmart built for AI agent shopping?

Walmart has combined a distinct agent identity, customer-defined purchase controls and a dedicated route for agents to interact with its retail systems.

Agent Profiles will sit inside a customer’s Walmart account, but separate an agent’s activity from the customer’s personal shopping. Customers will be able to review purchases by agent and decide whether each purchase needs approval. By default, every agent purchase requires customer approval.

Customers may instead set a spending limit for an agent, with Walmart giving examples of $50, $100 or $200. This permits automatic processing within the customer’s chosen limit, but it is not an unconditional authorisation: Walmart may require additional approval even for a purchase that falls within it.

The design resembles established authorisation controls in payments. An issuer can distinguish a cardholder or authorised user, apply a limit and monitor activity. Walmart is applying a similar control logic to a software agent inside a merchant account.

Why is Walmart separating agent identity from the customer account?

A separate profile makes it easier to identify which purchases an agent made and to manage that agent’s permissions without blending its activity into a person’s shopping history.

Without such separation, an agent using a customer’s account could make purchases that appear alongside the customer’s own activity. A dedicated profile gives the merchant and customer a clearer boundary: the agent is acting under delegated authority, not simply impersonating the account holder.

For payment operators, this is a meaningful move from informal delegation towards auditable delegation. An agent profile can become a point where permissions, activity and spending rules are associated with a particular agent. That creates a more practical foundation for monitoring and dispute review than a shared account alone.

It also distinguishes Walmart’s announcement from developments focused primarily on agent identity protocols or payment-network infrastructure. Our earlier analysis explored what agentic commerce looks like when it goes live. Walmart’s focus is the merchant’s own account, acceptance rules and controls.

How do spending limits and customer approvals change the risk model?

Spending limits place a clear boundary around what an agent can do, while default approval leaves the customer in the decision loop for each purchase.

The $50, $100 and $200 examples are not merely interface settings. They illustrate how merchants can make delegated purchasing more specific than “the agent may shop”. A limit creates a defined level of authority for a particular agent, and separate profiles allow customers to manage agents independently.

The model is deliberately not a blanket commitment to autonomous checkout. Walmart retains the ability to require further approval. This creates a layered control structure: customer settings define the starting permission, while merchant rules can add another checkpoint.

Payment providers should consider how comparable rules could align with existing authorisation and risk systems. Per-agent spending limits could be designed in ways analogous to velocity controls, though Walmart’s announcement does not specify a full technical implementation. The operational task is to decide how an agent’s identity, permission and transaction history should inform the provider’s existing controls.

Why does keeping card data away from agents matter?

Walmart says agents cannot access, add or change a customer’s payment card information, and its updated terms prevent agents from receiving payment-card information.

This is a significant boundary between shopping and payment credentials. An agent can help select or purchase an item without being handed the card details that fund the transaction. In other words, the payment credential stays with the human-controlled account rather than becoming part of the agent’s accessible data.

That boundary supports a safer approach to PCI compliance, but it should not be mistaken for proof of compliance by itself. Providers still need to assess their own systems, data flows and responsibilities. Payment tokenisation may be relevant to how credentials are protected in an implementation, but Walmart’s announcement does not say that its Agent Profiles use a specific tokenisation design.

The principle for payments companies is more general: keep payment credentials outside the agent’s scope wherever possible, and give the agent only the data needed for its task. Walmart also says standard returns processes apply to agent purchases, extending an existing customer protection to this new shopping channel.

What does the Walmart Agent Gateway signal to payment providers?

The Walmart Agent Gateway signals that large merchants are preparing a machine-readable front door for shopping agents, rather than relying only on interfaces designed for people.

The developer beta is built on the open Model Context Protocol (MCP). Walmart describes the Gateway as a way for external agents to shop without sharing customers’ personal information or broader purchase history. The company says agents may be helped to find usual items without receiving that wider information.

Walmart has already integrated OpenAI’s ChatGPT, Google Gemini and its own Sparky assistant. The Gateway is a further step towards connecting external agents to shopping functions through an agent-oriented interface.

For POS vendors, PayFacs and other payment providers, MCP is a signal to examine how product and checkout information can be presented in machine-readable form. This does not mean every provider needs to adopt MCP immediately. It does mean that product availability, fulfilment options and purchase rules may need to be legible to software agents, not just displayed for human browsing.

The Walmart Agent Gateway is in developer beta. Its development sits within a broader strategic contrast: Walmart is inviting agent shopping on defined terms, while Amazon has blocked external agents and sued Perplexity over its Comet browser agent. TechCrunch has also framed Meta’s Muse agent as facing resistance.

What should merchants and payment providers do now?

Merchants and payment providers should define their agent acceptance policy before agents become a routine route to checkout.

First, decide which agents or agent providers may interact with your service, and what identity or verification is required. Second, define the actions an agent may take and whether each requires customer approval. Third, consider per-agent limits in the context of existing velocity, fraud and authorisation controls. Fourth, keep payment-card data out of agent permissions. Finally, prepare product catalogues and checkout information for machine-readable access.

These are retail payment strategy decisions, not simply technical integrations. A merchant that welcomes agents without defining boundaries may create uncertainty about what was authorised, which agent acted and how a purchase should be reviewed. A provider that plans those controls early can help customers benefit from automation without making payment credentials or account access unnecessarily broad.

This discussion is separate from familiar checkout questions such as Walmart Apple Pay. The central issue here is not which payment option a shopper selects, but how an AI agent receives authority to shop and where the boundaries of that authority sit.

Key Takeaway

Walmart is treating agent shopping as a permissioned merchant channel, not an unrestricted extension of a customer login. Separate agent profiles, default purchase approval, optional spending limits and exclusion of card data establish a practical control plane for AI agent payments. Payment providers should use this development to prepare their own policies for agent identity, bounded authority and machine-readable checkout.

Frequently asked questions about Walmart’s AI agent shopping controls

What are Walmart Agent Profiles?

Agent Profiles are dedicated sub-profiles inside a customer’s Walmart account. They separate an AI agent’s shopping activity from the customer’s personal purchases.

Can a Walmart AI agent buy items without approval?

By default, customers must approve every agent purchase. They can opt into a spending limit for an agent, but Walmart may still require additional approval.

Can AI agents access Walmart customers’ payment card details?

Walmart says agents cannot access, add or change customers’ payment card information. Its updated terms also prevent agents from receiving payment-card information.

What is the Walmart Agent Gateway?

It is a developer beta that connects external AI agents to Walmart’s shopping capabilities using the open Model Context Protocol. Walmart says it is designed to let agents shop without sharing a customer’s personal information or broader purchase history.

Why should payment providers care about Walmart’s agent controls?

The design demonstrates how a merchant can separate agent identity from customer identity, constrain spending authority and keep card data outside an agent’s scope. These are relevant considerations for providers preparing their own agent acceptance and checkout policies.