Walmart has opened its online store to outside AI shopping agents, and it has written the rules those agents must follow. On Wednesday, October 21, 2026, the retailer announced three linked pieces: Agent Profiles inside customer accounts, a Walmart Agent Gateway built on the Model Context Protocol (MCP), and a published set of terms for AI agents. Chief executive John Furner discussed the plan the same day at the Masters of Scale Summit in San Francisco, according to TechCrunch.
The move puts the two largest US online retailers on opposite sides of the agentic commerce question. Amazon has spent the autumn blocking third-party agents, including Meta’s Muse, from transacting on Amazon.com. Walmart is doing the reverse: inviting approved agents in through a front door it controls, with customer-set spending limits, default purchase approval and a promise that agents never see card data. Reuters, TechCrunch, Retail Dive, Business Insider and The Verge all reported the launch within hours.
In short
- Agent Profiles arrive “in the next month”: each AI agent gets its own profile in a Walmart account, with purchases shown separately and either per-order approval or a spending limit (Walmart’s examples are $50, $100 and $200).
- Walmart Agent Gateway is an MCP server that lets approved agents search Walmart and Sam’s Club inventory and place orders; it starts as an invitation-only beta for companies and developers.
- Terms for AI agents bar agents from receiving payment card data, changing payment methods or acting beyond the shopper’s intent; non-compliant agents “will be blocked from Walmart’s site.”
- Returns still apply, including when an agent “doesn’t act as intended,” and no Walmart+ membership is required to use agent shopping.
- The Amazon contrast is the story: Amazon blocked Meta’s Muse on September 20, while Walmart already lists Muse, ChatGPT and Gemini among its conversational commerce integrations.
What exactly did Walmart announce on October 21?
Walmart’s newsroom post, dated October 21 and published without a named author, describes three components that work together. The company framed the goal plainly: “Together, these tools will let customers shop with Walmart and Sam’s Club through the apps and agents they choose.” It also set expectations low on adoption, adding that “shopping with an agent won’t be for everyone, and it doesn’t have to be.”
The first component is a consumer feature, Agent Profiles, which changes what a Walmart account looks like. The second is infrastructure, the Agent Gateway, which changes how an agent talks to Walmart’s systems. The third is a contract, the terms for AI agents, which sets what any agent may and may not do on Walmart’s properties regardless of which route it uses.
Furner presented the technology at the Masters of Scale Summit in San Francisco on the day of the announcement. TechCrunch and Techstrong.ai both paraphrased his remarks rather than quoting him directly: the gist, per Techstrong, was that Walmart is preparing for a future in which consumers routinely rely on autonomous software to find deals and make purchases.
Which agents are already connected?
Walmart’s post lists existing conversational commerce integrations with OpenAI’s ChatGPT, Google’s Gemini and Meta’s Muse, according to TechCrunch and PPC Land. TechCrunch had earlier reported that Walmart partnered with Muse at Meta’s Connect developer conference in September, and that Walmart described some earlier blocking of Muse on its site as unintentional.
The Gateway itself, however, opens as an invitation-only beta. Walmart has not published the list of agent companies admitted to it, which PPC Land flagged as one of the main open questions. The beta page also warns that access “may be suspended at any time.”
How do Walmart Agent Profiles work?
An Agent Profile is a dedicated sub-identity for a specific AI agent inside a customer’s Walmart account. Purchases the agent makes appear under that profile, separate from the customer’s own orders, so a household can see at a glance which items a bot bought. Walmart compares the setup to “adding someone as a cardholder on your credit card,” per PPC Land’s reading of the newsroom post.
The profile is created automatically the first time a customer uses a given agent through the Agent Gateway. The customer then links it by signing in to Walmart, chooses controls and limits, and can delete the profile at any time. TechCrunch reports that any Walmart account holder can use the feature: a Walmart+ subscription is not required.
Walmart says profiles will roll out “in the next month,” which most outlets read as November 2026. No exact date has been given, and PPC Land notes it is not yet clear whether Sam’s Club accounts get the feature on the same day as Walmart accounts.
Approval by default, or a spending limit
Customers choose between two approval models. The default is approval for every purchase: the agent assembles a cart, and the human signs off before money moves. The alternative is a spending limit the customer sets, with Walmart citing $50, $100 and $200 as examples.
Two details remain undefined. Walmart has not said whether a limit applies per order, per day or per month, and it has not said whether different agents can carry different limits on the same account. The terms also reserve Walmart’s right to “require additional approvals even for purchases within an agent’s spending limits,” which gives the retailer a manual brake on unusual orders.
| Control | What Walmart says | Still unclear |
|---|---|---|
| Profile creation | Created on first use of an agent through the Gateway; linked when the customer signs in | Whether browser-based agents outside the Gateway get profiles |
| Default approval | Every purchase needs customer approval | How approval prompts reach the customer (app, email, agent interface) |
| Spending limit | Customer-set; examples of $50, $100, $200 | Per order or per period; per agent or per account |
| Extra checks | Walmart may require additional approvals within limits | Which triggers apply |
| Revocation | Profiles can be deleted at any time | Whether pending orders are cancelled on deletion |
| Eligibility | Any Walmart account; Walmart+ not required | Sam’s Club timing |
What is the Walmart Agent Gateway?
The Agent Gateway is Walmart’s own MCP server: a structured, machine-readable entry point that approved agents call instead of scraping web pages. MCP is an open standard, originally created by Anthropic, that lets AI systems connect to external data, tools and workflows. Through the Gateway an agent can search Walmart’s online inventory and place an order on the customer’s behalf, according to TechCrunch.
Walmart’s beta page, as described by PPC Land, lists the selling points it is pitching to agent developers: no website scraping, lower latency, lower token usage, “verified identities,” and “category restrictions, rate limits & compliance guardrails.” The same page says the Gateway “connects AI agents through Sparky,” Walmart’s in-house assistant, a detail the newsroom post does not repeat.
A key privacy design point is that a customer can buy their usual brands without handing their purchase history to the agent. Walmart keeps the history on its side and answers the agent’s request, so a third-party model never needs the full order record to reorder a regular detergent.
Why route agents through a gateway at all?
For Walmart, a gateway turns an unpredictable stream of browser automation into a known, authenticated channel. It can rate-limit traffic, restrict categories, and see which company’s agent sent each order. For the agent developer it is cheaper: a structured response consumes fewer tokens than rendering and parsing a full product page.
The industry data shows why identity matters. Arkose Labs reported on October 20 that only 2.3% of legitimate agent sessions it observed carried verified Web Bot Auth signatures, per PPC Land. HUMAN Security said in September that Muse alone accounted for 72% of the daily AI agent requests it observed. A retailer that cannot tell good agents from scrapers has two blunt choices: block everything, as many sites have, or build a door that only identified agents can use, which is Walmart’s route.
That is also the practical lesson of our explainer on what a store must enable before agents can buy: the checkout plumbing, not the chatbot, decides whether agent orders actually complete.
What do Walmart’s terms for AI agents require?
The terms are short and written in plain language. PPC Land counted eight points in the newsroom summary, and Walmart’s beta page links to a fuller “Agentic terms” document that has not yet been widely reviewed. Neither the effective date nor the enforcement mechanism has been published.
- Customers stay in control, and Walmart’s returns process applies, “including if an agent doesn’t act as intended.”
- Approval by default: purchases need customer approval unless the customer sets a spending limit, and Walmart may still ask for more.
- Intent boundary: agents may not take actions beyond what the shopper intends.
- No card data: agents do not receive payment card information and cannot add or change payment methods.
- Data minimization: agents get only the information needed to complete the task.
- Separate privacy policies: each party’s privacy policy covers the data it handles.
- Walmart’s Privacy Notice governs data an agent provides to Walmart.
- Enforcement: “Agents that do not follow these terms will be blocked from Walmart’s site.”
Retail Dive’s reading emphasized the first and fourth points: customers must approve agent actions, and agents never touch the card. TechCrunch highlighted the returns guarantee, which addresses the most obvious consumer fear about delegating a purchase: receiving the wrong item because a model misread the instruction.
The returns clause shifts risk onto Walmart
Accepting returns when an agent misbehaves means Walmart, not the customer and not the agent developer, absorbs the cost of model error at the point of sale. That is a commercial bet that agent-driven orders will be accurate enough, and incremental enough, to justify the occasional reverse-logistics bill. It also gives Walmart an incentive to police agent quality through the Gateway.
It does not settle the payments side. Card networks have not finalized how disputes work when an authorized agent buys something the cardholder later disowns, an issue we covered in our analysis of why agent chargeback rules likely miss the 2026 holidays. Walmart’s design avoids part of that problem by keeping card data inside its own wallet, so the agent never initiates a card transaction directly.
How does Walmart’s approach compare with Amazon’s?
The contrast is stark and deliberate. Amazon blocked Meta’s Muse from transacting on Amazon.com on the night of September 20, 2026, telling users that “continued access by an unauthorized AI agent violates Amazon’s Conditions of Use.” Amazon objected that Meta had not told it Muse would access the store, that the agent did not identify itself as automated, and that Muse appeared to capture and store customer credentials. Our earlier report covers how Amazon blocked Meta’s Muse while Shopify opened Shop Pay checkout instead.
Amazon’s legal footing had narrowed before that block. In August 2026 the Ninth Circuit vacated the injunction that had kept Perplexity’s Comet shopping agent out of Amazon accounts, finding that the user, not the developer, does the accessing; it denied rehearing on September 10. That leaves Amazon its terms of service, rate limiting and technical blocking, which it is now using.
Retail analysts quoted by Techstrong.ai suggested Amazon also has commercial motives: protecting its advertising business, which depends on human shoppers seeing sponsored placements, and steering customers to its own assistant, Alexa for Shopping. Walmart’s open stance instead fits its everyday low price model, since an agent programmed to find the lowest price for a basket of staples is, in theory, a natural Walmart customer.
| Question | Walmart (Oct 21, 2026) | Amazon (as of Oct 2026) |
|---|---|---|
| Third-party agents allowed to buy? | Yes, approved agents via Agent Gateway | No; Muse blocked Sept 20, Perplexity Comet contested |
| Agent channel | MCP server (invitation-only beta) | None for outside agents |
| Customer controls | Agent Profiles, per-order approval or spending limit | Not applicable to outside agents |
| Card data shared with agent | Never, per terms | Not applicable |
| In-house assistant | Sparky (Gateway connects through it, per beta page) | Alexa for Shopping |
| Existing agent partners | ChatGPT, Gemini, Meta Muse | None announced for outside agents |
| Enforcement tool | Block agents that break terms | Conditions of Use, technical blocks, litigation |
Walmart and Amazon are not the only reference points. Shopify announced within a day of the Muse block that Muse would be able to complete purchases through Shop Pay across Shopify stores. At the other end, TechCrunch reported on October 6 that early adopters had complained on social media about agents being blocked by brands and sites including Yelp, eBay, Zillow, Pizza Hut, Adidas and several airlines.
Walmart’s position is therefore not simply “open.” It is open on Walmart’s terms, through Walmart’s server, with Walmart’s identity checks, and with the retailer’s own Sparky assistant reportedly sitting in the path. That is closer to the model we described in our forecast that agentic commerce settles on retailer-controlled checkout than to an unrestricted web.
Why is Walmart doing this now?
Walmart has been building toward an agent channel for a year, and the October announcement is the first time it has published the governance layer. The timeline matters because the earlier steps did not all go well, and the new design reads like a response to those lessons.
- September 29, 2025: OpenAI launched Instant Checkout in ChatGPT with Stripe, with Walmart among the launch partners, per PPC Land.
- January 11, 2026: Google published the Universal Commerce Protocol (UCP) with Walmart among its co-developers; Walmart and Target announced checkout inside Gemini the same day.
- March 2026: a Walmart executive said purchases completed inside ChatGPT converted three times worse than click-outs to Walmart’s own site, and OpenAI shut down Instant Checkout that month. Walmart moved Sparky into ChatGPT and later Gemini.
- September 2026: Walmart partnered with Meta’s Muse, announced at Meta Connect.
- October 21, 2026: Agent Profiles, the Agent Gateway and agent terms are announced.
The March data point is the most revealing. If purchases completed inside a third-party chat window convert worse than sending the shopper to Walmart, the retailer has every reason to keep the transaction on its own systems. The Gateway does exactly that: the agent can be anywhere, but the cart, the payment and the account all stay at Walmart.
Timing also lines up with the holiday season. Profiles are due within a month, which would put them live before the peak November and December shopping weeks if Walmart hits its schedule. Adobe has forecast 130% growth in AI referral traffic for the 2026 holidays, a figure we have argued is likely to come in below forecast; either way, a retailer that offers agents a clean channel before the season is better placed to capture whatever volume arrives.
Walmart is also competing on membership. Days earlier it bundled Walmart+ with Sam’s Club Plus at $120 a year, a pitch aimed squarely at Prime households. Making agent shopping available without a Walmart+ subscription widens the funnel rather than gating it.
What does this mean for pricing and the surveillance pricing debate?
Walmart used the announcement to restate a pricing pledge: “We price products, not people,” with the company saying the prices an AI agent sees are the same prices shown to human shoppers. That line answers a live controversy. Lindsay Owens, chief executive of the advocacy group Groundwork Collaborative, has accused Walmart of using dynamic AI software to adjust prices based on individual behavior, according to Techstrong.ai, and Walmart rejected the claim on Wednesday.
The dispute has regulatory weight. Groundwork published a review of Walmart pricing patents in early October, and Walmart responded with an open letter and said it would not renew one cart-aware shelf label patent at the center of the report. At the same time, the Federal Trade Commission is preparing compulsory 6(b) orders on personalized pricing, a step covered in our report that the FTC is preparing personalized pricing orders with delivery and rideshare first.
Agents complicate that debate in two directions. An agent comparing prices across retailers makes individualized pricing easier to detect, because the same request from different accounts should return the same number. But an agent channel also generates new behavioral data, including which agent a shopper uses and what limits they set, that a retailer could in principle use. Walmart’s terms restrict what agents receive; they say less about what Walmart learns from agent traffic, beyond placing agent-provided data under its Privacy Notice.
PPC Land raised a related, unanswered question: how sponsored listings and search ranking apply when an agent, not a person, reads the results. Walmart Connect, the retailer’s ad business, outlined AI plans in January 2026 that included ad tests in Sparky. If the Gateway routes agents through Sparky, the commercial logic of Walmart’s ranking could follow the request into the agent’s answer, and Walmart has not yet said how it will label or limit that.
What does it mean for brands and Walmart Marketplace sellers?
For suppliers and third-party sellers, the most immediate change is a new kind of shopper. An agent with a $100 limit and an instruction to restock household basics will not browse a brand’s hero imagery or read lifestyle copy. It will read structured attributes: size, count, unit price, availability, delivery date and ratings.
That shifts emphasis toward catalog hygiene. Accurate unit pricing, complete attribute fields, consistent variant naming and in-stock signals become the basis on which an agent picks one listing over another. Brands that already optimize for search on Walmart.com will find much of that work transfers, but descriptive marketing copy loses value relative to structured data.
Category restrictions and rate limits
The beta page’s mention of “category restrictions” deserves attention from sellers in regulated or age-restricted categories. Walmart has not said which categories agents cannot buy, but alcohol, tobacco, firearms-related goods, some pharmacy items and age-gated products are the obvious candidates. Sellers in those categories should assume agent volume will be capped or excluded until Walmart says otherwise.
Rate limits matter to sellers too. An agent cannot hammer the Gateway to watch for a price drop or a restock the way a scraper might, which reduces the advantage of automated deal-hunting tools and the load they put on listings. For Marketplace sellers that compete on price, it also means agent shoppers will see prices at the cadence Walmart allows, not in real time.
| Group | What changes | Immediate action |
|---|---|---|
| Brands and suppliers | Agents choose on structured attributes and price | Audit item attributes, unit pricing and availability data |
| Marketplace sellers | Agent orders arrive with spending caps attached | Check how fees, shipping and bundles render in a capped basket |
| Agent developers | Approved access via MCP replaces scraping | Apply to the invitation-only beta; review the agent terms |
| Rival retailers | A large retailer now publishes an agent rulebook | Decide between blocking, a gateway, or an open protocol |
| Shoppers | Separate agent purchase history, limits and returns | Start with per-order approval; set limits only for routine reorders |
How many shoppers will actually use AI agents to buy?
Walmart itself hedged on adoption, saying agent shopping “won’t be for everyone.” The available survey data supports caution. A Coveo report cited by Techstrong.ai found that only 16% of shoppers currently trust AI assistants to complete transactions on their behalf.
Where trust exists, it concentrates in routine, utilitarian purchases. Consumer research cited in the same coverage suggests shoppers are more willing to delegate household goods and groceries than discretionary items, and those are categories Walmart already dominates. That makes Walmart’s bet less about a general shift in behavior and more about capturing replenishment orders, which are high frequency, low emotion and well suited to a spending cap.
Meta’s Muse gives a sense of the potential reach. CNBC, citing Sensor Tower, reported that Muse had about 2.5 million downloads within roughly two weeks of its September 8 launch and reached the top of the US free iOS chart. If HUMAN Security’s figure that Muse generated 72% of observed daily agent requests holds, Walmart’s existing Muse partnership positions it to receive much of the agent traffic that Amazon is turning away.
Retailers will struggle to measure agent shopping cleanly for some time. Verified bot identity is rare, per Arkose’s 2.3% figure, and agent sessions that come through a browser look like human traffic in standard analytics. Walmart’s Agent Profiles solve the attribution problem for Walmart, since every agent order is tagged to a profile, but suppliers will only see those numbers if Walmart chooses to report them.
What questions remain unanswered?
Walmart’s announcement is a framework, not a finished product, and several important details are missing. PPC Land listed the most consequential gaps, and the first weeks of the beta will show how Walmart fills them.
- Who is in the beta: Walmart has not named the agent companies with Gateway access.
- Limit mechanics: whether spending limits apply per order or per period, and whether they vary by agent.
- Effective date of the terms: the “Agentic terms” document has no published start date.
- Browser agents: whether agents that drive a browser on Walmart.com must migrate to the Gateway, and how non-compliant agents will be detected and blocked.
- Restricted categories: which product categories agents cannot buy.
- Sam’s Club timing: whether Sam’s Club members get Agent Profiles at the same time as Walmart customers.
- Ads and ranking: how sponsored listings appear in results returned to agents.
The protocol question is also open. Walmart co-developed Google’s UCP, yet the October 21 post centers on MCP and does not mention UCP. Whether the Gateway eventually speaks both, or whether Walmart treats MCP as the transport and UCP as the checkout layer, will matter to developers deciding which standard to build against.
The near-term markers are concrete. Agent Profiles are due within a month; the beta roster will show which agent companies Walmart trusts; and Amazon’s response, whether it opens its own sanctioned channel or keeps blocking, will define the competitive landscape for the 2026 holidays. Regulators are watching the pricing side, with the FTC’s personalized pricing work and state laws in Maryland and Connecticut already in force.
Frequently asked questions
What did Walmart announce about AI agents?
On October 21, 2026, Walmart announced Agent Profiles, the Walmart Agent Gateway and terms for AI agents. Together they let approved third-party AI agents search Walmart and Sam’s Club inventory and buy on a customer’s behalf, under customer-set controls.
When will Walmart Agent Profiles be available?
Walmart says Agent Profiles will roll out “in the next month,” which most coverage reads as November 2026. No exact date has been published.
Do I need Walmart+ to let an AI agent shop at Walmart?
No. According to TechCrunch, any Walmart account holder can use agent shopping, and a Walmart+ membership is not required.
Can an AI agent see my credit card details at Walmart?
No. Walmart’s terms say agents do not receive payment card information and cannot add or change payment methods. Payment stays inside the customer’s Walmart account.
How do spending limits for AI agents work?
Each agent gets its own profile, and the customer chooses either approval for every purchase (the default) or a spending limit, with Walmart citing $50, $100 and $200 as examples. Walmart has not said whether limits apply per order or per period, and it may still ask for extra approval within a limit.
What happens if an AI agent buys the wrong item?
Walmart’s terms say its standard returns process applies, including when an agent “doesn’t act as intended.” Agent purchases are also shown separately in the account, which makes them easier to identify.
What is the Walmart Agent Gateway?
It is Walmart’s MCP server, a structured channel that lets approved AI agents search inventory and place orders without scraping web pages. It launches as an invitation-only beta for companies and developers.
Which AI agents work with Walmart?
Walmart lists existing integrations with OpenAI’s ChatGPT, Google’s Gemini and Meta’s Muse. The companies admitted to the new Agent Gateway beta have not been named.
Why does Amazon block AI shopping agents while Walmart allows them?
Amazon blocked Meta’s Muse on September 20, 2026, citing undisclosed access, a lack of agent self-identification and credential handling. Analysts also point to Amazon’s advertising business and its own Alexa for Shopping assistant. Walmart’s low-price positioning suits agents that hunt for the cheapest basket, so it has chosen a controlled gateway instead.