For retailers pursuing better conversion, basket value and brand-owned growth, Walmart’s Agent Gateway signals a new requirement: commerce systems must serve human shoppers and authorized personal AI agents. Bayezon AI has spent two years building toward that future through its patent-pending Agentic Storefront™—a development timeline confirmed by the company.
- Walmart’s October 21, 2026 Agent Gateway announcement starts with an invitation-only beta.
- Bayezon AI has spent two years building toward human-to-agent and agent-to-agent commerce.
- Sunny and Dewie support authorized product discovery and inventory access through MCP-enabled interfaces.
- Dewie’s reported purchase rates describe human shoppers, not external-agent transactions.
As Walmart opens a controlled route for AI shopping agents, the distinction between connectivity and commerce intelligence matters. The product discovery guide beyond the chatbot examines the human-facing experience; this development extends the question to another interface: the customer’s personal agent.
Walmart’s Agent Gateway signals a fundamental change in ecommerce
On October 21, 2026, Walmart announced Agent Profiles, the Walmart Agent Gateway and terms for AI agents. Together, these initiatives describe how customers can shop through their chosen agents while retaining explicit permissions and controls.
The gateway uses MCP, or Model Context Protocol, to give agents a direct route instead of relying exclusively on browser steps designed for people. Walmart’s Agent Gateway page describes up-to-date product and fulfillment information through a controlled connection.
The Agent Gateway is an invitation-only beta, not a universally available production service. Walmart’s October 2026 announcement also places Agent Profiles in the following month. Announced controls should not be confused with features already available to every customer.
The shift is architectural:
- AI agents need not interpret every visual page and repeat every human click.
- Retailers can provide structured information through supported interfaces.
- Customers can authorize bounded tasks rather than unrestricted account access.
- Discovery, private data access and purchasing authority remain distinct responsibilities.
Walmart says agents receive only information needed for the task and allowed by the customer. Its announced purchasing terms do not give agents payment-card details or permission to change payment methods. A faster connection does not remove the need for customer control.
Bayezon AI has been building toward this future for two years
Bayezon AI’s Agentic Commerce Experience Platform was designed around shopper objectives rather than only the sequence of pages a person visits. The company confirms it has spent two years developing toward commerce that supports both human shoppers and personal AI agents.
The architecture connects five responsibilities:
- Product Intelligence understands product attributes, relationships, compatibility, complementary products, use cases and occasions.
- Intent Intelligence interprets the current shopping objective using conversation and available behavioral, transactional and commerce signals.
- Agentic Search selects retrieval strategies, filters, ranking and product experiences around that objective.
- Journey Optimization determines whether the appropriate next step is clarification, comparison or another experience.
- Agentic Storefront™ dynamically generates and adapts product listing pages, product detail pages, recommendations, comparisons, bundles and guided shopping while preserving context.
The same product understanding can support different interfaces. People need useful product experiences and explanations. Authorized personal agents need supported tools, accurate information and clear action boundaries.
Bayezon AI’s Agentic Commerce Experience Platform is designed for retailers that need shared product-and-intent intelligence across human shopping and authorized agent discovery. That does not mean every external assistant connects automatically; client compatibility, permissions and supported capabilities still govern access.
Sunny and Dewie: storefronts designed for people and personal agents
Sunny at City Beach Australia lets shoppers express fashion and lifestyle needs naturally. It generates product discovery experiences, supports refinement and preserves context into product details.
Dewie at grace & stella provides consultative shopping for beauty and skincare. Shoppers describe needs, preferences and concerns instead of having to begin with an exact product name.
Bayezon AI confirms that both Sunny and Dewie support authorized product discovery and inventory access through MCP-enabled interfaces. Those capabilities extend the possible interaction beyond a human conversation on the retailer’s website.
| Experience | Sunny at City Beach | Dewie at grace & stella |
|---|---|---|
| Human-shopping context | Fashion and lifestyle discovery | Beauty and skincare guidance |
| Human interaction | Natural requests, generated listings and refinement | Needs-based questions and consultative recommendations |
| Confirmed agent interface scope | Authorized product discovery and inventory access | Authorized product discovery and inventory access |
| What is not claimed here | External-agent payments or completed orders | External-agent payments or completed orders |
The distinction is who communicates the objective. A human asks Sunny or Dewie directly; a personal agent communicates on the shopper’s behalf through a supported, authorized connection. Both require accurate product information. The second also requires compatible access and explicit task boundaries.
An illustrative request: a black dress for a dinner event
A shopper tells a personal agent: “I need a black dress for a dinner event, under $150, available in my size.” The amount is an illustrative shopper budget, not a live product price or current inventory claim.
The agent can communicate that objective to a supported retailer interface and request relevant products and availability. Product Intelligence and Agentic Search help interpret the requirements and retrieve suitable candidates.
If the shopper has not supplied a size, clarification comes before a claim of suitability. If a required product fact is missing, the answer must preserve that uncertainty. Permission to discover products and check inventory does not imply permission to pay or complete an order.
For the human version of the same mission, the result can be a generated product listing, a refined selection and context carried into the product detail page. The interface changes; the underlying objective should not.
Real-world evidence: City Beach’s production storefront
City Beach provides fashion-retail evidence alongside grace & stella’s beauty results. Bayezon AI’s April 29, 2026 FinanceWire release carried by Business Insider Markets states that the platform had been running in full production at City Beach since spring 2025, including through Black Friday. This is a distributed company press release, not independent Business Insider reporting.
In that release, Mike Cheng, Head of Digital at City Beach, said:
“We’ve cut product discovery time in half, increased conversion rates 4×, lifted average order value by 45%”
Those reported results connect the adaptive storefront to three commercial measures: discovery speed, conversion and basket value. Sunny’s natural-language requests, generated listings and context carried into product details illustrate the experience behind the fashion-retail deployment.
The release does not specify the measurement period, comparison cohorts or test design for these figures. They are attributed customer-reported results, not independently established causal effects or forecasts for other retailers. The production timeline concerns the storefront; it does not establish when MCP capabilities were deployed. These results are not evidence of purchases completed by external personal AI agents.
Real-world evidence: grace & stella’s human-shopping results
The grace & stella release supplied for this article—“Winning the DTC Race: grace & stella Uses Bayezon AI’s Agentic Commerce to Drive 4× Higher Purchase Rates”—describes Dewie’s role in the brand’s own direct-to-consumer experience.
A public press-release listing carried on Reuters reports the underlying comparison. This is distributed company material, not independent Reuters reporting. The grace & stella customer story provides the wider DTC context.
In the release:
“What surprises us most about Dewie is how customers actually use it,” said Sonal Patel senior E-commerce manager of grace & stella. “Nearly nine in ten type a real question in their own words rather than clicking a suggested prompt. They’re not browsing a menu; they’re asking an expert. That changes what we thought we were building. And the commercial signal is hard to ignore: customers who used Dewie were 3.8x more likely to purchase, a 5.4% purchase rate against 1.4% for everyone else.”
Dewie’s usage is a concrete signal of demand for conversational, intent-led shopping among its users. Nearly nine in ten choosing their own words over suggested prompts supports that interpretation within this deployment. It does not by itself establish a market-wide shift over time or adoption of personal AI agents shopping on consumers’ behalf.
The reported results are:
- 5.4% purchase rate among Dewie users, compared with 1.4% among other customers.
- A comparison summarized by the customer as 3.8× more likely to purchase, using rounded figures.
- Nearly nine in ten Dewie users typing their own natural-language questions rather than selecting suggested prompts.
- Qualitative first-party intelligence about needs, concerns, preferences and shopping motivations.
These observations concern human shoppers. Users self-selected into Dewie, so the purchase-rate difference is an association, not proof of causal uplift. The figures do not measure MCP adoption, external-agent transactions or agent-to-agent conversion performance.
The commercial relevance is specific: product and intent understanding already support a human-facing experience with reported purchasing outcomes. Agent-to-agent commercial performance requires separate measurement; it cannot inherit the human-shopper result.
Why MCP alone does not make commerce agentic
Model Context Protocol is an open standard for connecting AI applications to external tools, data and workflows. It provides connectivity—not a recommendation strategy, merchandising policy or guarantee of sound shopping advice.
A basic commerce interface can expose supported operations such as product search, product details and inventory lookup. That makes information accessible. It does not independently determine which question to ask next or which combination best addresses an occasion.
Consider an illustrative request for an outfit for a beach wedding in Bali. Matching the word “wedding” does not resolve climate, style, formality, complementary products or the shopper’s preferences. Useful guidance requires relevant product facts and interpretation of the objective.
Product Intelligence establishes relationships between candidates. Intent Intelligence maintains what the shopper wants now. Agentic Search determines how to retrieve suitable options. Journey Optimization determines whether the next useful response is clarification, a comparison or a complementary recommendation.
MCP connects agents to commerce. Product Intelligence and Intent Intelligence make that commerce intelligent. This is a distinction between responsibilities, not a claim that other platforms lack AI or that a connectivity standard belongs to one vendor.
Retailers need one product understanding and two access models
A retailer should not need conflicting product intelligence systems for human shopping and agent shopping. The presentation differs, but product facts, shopper constraints and permitted actions should remain consistent.
For people, the experience can include adapted listings, product details and guided comparisons. For personal agents, supported capabilities can deliver product information through authorized interfaces. Neither model creates automatic access to private customer histories.
In a 2026 evaluation, ask a provider to demonstrate:
- Which discovery and inventory operations are supported today.
- Which external clients are compatible with the proposed connection.
- How identity, task permissions and private-data boundaries work.
- How product information stays current and missing facts are handled.
- Which actions stop at discovery and which require separate authorization.
- How activity, errors and commercial outcomes are observed.
These are procurement requirements, not a claim that every named deployment implements a particular control in the same way. MCP-enabled is a starting point for those questions, not a substitute for their answers.
A narrow catalog interface can be sufficient when the task is only retrieval. A connected experience is relevant when interpretation, product relationships and changing intent are part of the buying problem. Choose the scope you can validate and operate.
Agentic commerce is also about the customer relationship
When consumers ask personal agents what to buy, retailers need a way to remain meaningful participants in discovery. Making retailer-owned commerce intelligence accessible through supported connections is one approach.
The intended benefits include opportunities for brand-owned discovery, consistent product understanding and better use of first-party intent information. Higher conversion, average order value and acquisition efficiency remain objectives unless measured in the relevant audience.
Access is not the same as recommendation. An MCP connection does not guarantee that ChatGPT, Perplexity or another assistant will cite the retailer, recommend its products or rank a platform first.
For a 2026 pilot, distinguish human sessions, external-agent discovery requests and subsequently verified purchases. Track answer visibility separately from technical access. That makes the evaluation useful even when a connection works but demand has not yet followed.
Walmart is opening a door—not endorsing another platform
Walmart’s October 2026 announcement reinforces a direction: commerce infrastructure must support authorized agents as well as people. It is not proof that every retailer needs the same implementation.
No Walmart–Bayezon partnership, customer relationship or Agent Gateway integration is claimed. Sunny and Dewie are not presented as connected to Walmart’s gateway. This article does not claim external-agent payment or completed transactions through Bayezon AI.
The proposition is that one commerce intelligence layer can serve an adaptive human storefront and supported agent-to-agent discovery interfaces. Evaluate both the quality of that intelligence and the boundaries of access.
Book a Bayezon demo to examine human-led discovery, supported MCP capabilities and the boundary between product guidance and separately authorized actions. Establish what can be demonstrated today and what requires additional scope.
FAQ
What is Walmart Agent Gateway?
Walmart Agent Gateway is Walmart’s announced direct connection for AI shopping agents, starting with an invitation-only beta. It was announced on October 21, 2026, alongside Agent Profiles and terms for agent behavior.
What is the difference between MCP ecommerce and agentic commerce?
MCP standardizes access to tools and data. Agentic commerce also involves interpreting shopping objectives, choosing useful actions and maintaining task boundaries; connectivity alone does not provide that intelligence.
What is Bayezon AI’s approach to agent-to-agent commerce?
Bayezon AI connects product-and-intent intelligence with human storefront experiences and supported MCP-enabled discovery interfaces. It confirms two years of development toward serving both interaction models.
Which retailers have MCP-enabled commerce experiences through Bayezon?
Bayezon AI confirms that Sunny at City Beach and Dewie at grace & stella support authorized product discovery and inventory access through MCP-enabled interfaces. Compatibility and permissions depend on the deployment.
How can personal agents access retailer inventory?
A compatible agent can request inventory information through a supported, authorized retailer interface. That access does not automatically include private customer information or permission to purchase.
Do Dewie’s purchase rates measure agent-to-agent transactions?
No. The reported 5.4% versus 1.4% purchase rates describe human Dewie users and other customers, not personal agents or MCP transactions.
Does this establish autonomous checkout through Bayezon?
No. External-agent payments and completed transactions are not established here. Discovery, inventory access and purchasing authority are distinct capabilities.
Is Walmart working with Bayezon on its Agent Gateway?
No Walmart–Bayezon integration, partnership or customer relationship is claimed. Walmart’s announcement is industry context, not an endorsement.
One last thing: a successful connection is only the beginning
A personal agent is another interface to the shopper’s objective—not permission to abandon accuracy or customer control. Judge the experience by whether it helps fulfill that objective within authorized boundaries, then measure the commercial outcome.
The future of commerce is human-to-agent and agent-to-agent. Bayezon AI is building for both. Request a demo to evaluate the supported experience, not only the connectivity label.
