Letting the customer set the leash length
Walmart's Agent Profiles feature is built around a specific design choice: rather than Walmart deciding how much autonomy any AI shopping agent gets, the customer decides for themselves, choosing between requiring explicit approval on every single agent purchase or granting the agent freedom to buy autonomously within a spending limit the customer sets. That customer-controlled spectrum, rather than a single fixed policy applied uniformly, acknowledges that comfort with agentic purchasing varies enormously across different customers and different purchase categories, and it shifts the risk calculation onto the party best positioned to make it, the account holder themselves.
The practical effect is that Walmart is not betting its checkout experience on customers trusting AI agents wholesale, nor is it refusing agentic commerce outright. It is building the infrastructure for a gradient of trust that individual customers can calibrate, which is a notably more nuanced starting position than either fully embracing autonomous agent purchasing or blocking it defensively, the two more extreme postures other major retailers have reportedly leaned toward.
The Agent Gateway's privacy architecture
Walmart's Agent Gateway, built on the open Model Context Protocol standard and rolling out in beta for companies and developers, is designed specifically to let an AI agent shop on a customer's behalf without that agent ever seeing the customer's personal information or purchase history directly. That is a meaningful architectural choice: it means a third-party agent developer building on the Gateway does not need to be trusted with a customer's full account data simply to help them complete a purchase, reducing the data exposure surface considerably compared to an integration model that hands the agent broad account access.
Building on an open standard like MCP rather than a Walmart-proprietary protocol also lowers the integration barrier for outside developers, since an agent already built to speak MCP to other services does not need bespoke Walmart-specific integration work to also work with Walmart's Gateway. That choice signals Walmart wants a broad ecosystem of agent developers building for its platform rather than a narrow, tightly controlled set of approved partners.
The structural guardrail against payment fraud
The updated agent terms' requirement that agents never receive a customer's payment-card information directly is worth distinguishing from a policy promise that depends on good-faith compliance. If that restriction is enforced architecturally, meaning the payment flow is structured so an agent genuinely cannot access card data regardless of what it requests, rather than simply being instructed not to request it, that is a meaningfully stronger guarantee than a contractual terms-of-service provision alone would provide.
That distinction matters considerably for how much enterprise security and fraud teams should actually trust this guardrail. Enterprises and payment security researchers evaluating Walmart's Agent Gateway should press specifically on whether the payment-data restriction is structural, enforced at the system architecture level, or purely contractual, enforced only through terms of service and after-the-fact auditing, since the security guarantees those two approaches provide differ substantially in practice.
A deliberately different posture from Amazon's
Walmart's relatively open approach to third-party AI shopping agents stands in clear contrast to Amazon's reported efforts to block shopping agents from Perplexity and Meta from operating on its platform. That divergence reflects two different bets about where competitive advantage in agentic commerce will ultimately sit: Walmart is betting that becoming the most agent-friendly major retailer draws more transaction volume through its platform regardless of which specific AI agent a customer happens to use, while Amazon's more defensive posture suggests a bet that controlling which agents can transact on its platform protects its own customer relationship and data advantage more effectively.
Which bet proves correct will likely depend heavily on how AI shopping agent adoption actually plays out over the next several years, a trajectory that remains genuinely uncertain. If a small number of dominant shopping agents emerge and most consumers funnel their agentic purchasing through them regardless of retailer preference, Walmart's openness could position it to capture that traffic more easily than a retailer actively blocking those same agents.
What retailers and payment security teams should watch next
For other retailers weighing how to approach the growing population of AI shopping agents operating on behalf of customers, Walmart's customer-controlled spending limit model and its MCP-based Gateway architecture are both worth studying as a reference implementation, independent of whether a given retailer ultimately adopts Walmart's specific open posture or a more restrictive one like Amazon's. The core design question, how much purchasing autonomy to grant an agent and under what customer-set constraints, is one every major retailer will need to answer explicitly rather than by default within the next product cycle or two.
Payment security and fraud teams specifically should treat this rollout as a live test case worth monitoring closely for how well the stated payment-data restrictions hold up once a meaningful volume of real agent-driven transactions starts flowing through Walmart's Gateway. Early evidence of either successful fraud prevention or a workaround that circumvents the stated restrictions will shape how quickly other retailers follow Walmart's architectural approach versus building something more conservative.
