Walmart Agent Gateway vs Amazon's Agent Block: What CPG Brands Do Now

TL;DR

On October 21, 2026, Walmart announced Agent Profiles (spending limits and separate purchase history for each AI agent, launching within a month), the Walmart Agent Gateway (an MCP connection for agents, starting as a beta) and terms that agents must follow. Amazon has gone the other way and blocked Meta's Muse agent in September. For CPG brands this means AI agents will shop Walmart, Shopify stores and other open retailers through their catalog data, while on Amazon the agent that matters is Amazon's own Alexa for Shopping. Brands should get Walmart listings agent-readable, keep price, pack size and stock consistent across retailers, and keep optimizing Amazon listings for Amazon's assistant.

Walmart is building a front door for outside agents: Agent Profiles, an MCP-based Agent Gateway in beta, and published rules for agents.
Amazon blocked Meta's Muse about two weeks after its September 8 launch, so outside agents often can't see Amazon prices or listings at all.
Agents read structured catalog data, not your hero image. Walmart and Shopify listings with complete attributes are the ones an agent can compare.
Walmart's 'usuals' feature favors replenishment buying, which is most of CPG. Being the item an agent reorders is a new kind of loyalty.
On Amazon, keep working on listings for Alexa for Shopping, since that's the only agent Amazon is letting in.

On this page

What exactly did Walmart announce?
Why is Amazon keeping agents out?
What does this split mean for CPG brands?
Is any of this happening at real volume yet?
What should CPG brands fix first?
How do you know whether agents can find your products?
What happens next?
FAQ

Two of the biggest retailers in the country made opposite bets on AI shopping agents in the space of a month, and most CPG brands haven't noticed yet.

On October 21, Walmart announced three things: Agent Profiles, the Walmart Agent Gateway and a set of rules for AI agents. A few weeks earlier, Amazon had blocked Meta's Muse from shopping its store, roughly two weeks after Muse launched on September 8.

One retailer is building a front door for agents. The other is changing the locks. If you sell consumables on both, you now have two different problems to solve.

What exactly did Walmart announce?

Here's what's in Walmart's own post, stripped of the press-release warmth.

Agent Profiles give each AI agent a customer uses its own profile inside that customer's Walmart account. Today an agent just signs in as the customer, so its orders blend in with everything else. With a profile, the customer sees the agent's purchases separately, can require approval on every order or set a spending limit (Walmart's examples are $50, $100 or $200), and can delete the profile whenever they like. Walmart says these launch "in the next month."

The Walmart Agent Gateway lets agents shop through MCP, the Model Context Protocol, instead of clicking through a website built for humans. Walmart is candid that browser-driving agents are "slow and error-prone" and "may get stuck along the way." It starts as a beta for companies and developers. Reuters reported that assistants such as ChatGPT, Gemini and Muse will be able to connect to Walmart and Sam's Club systems through it, and that it won't require Walmart+.

Terms for AI agents spell out the rules. Customers approve purchases by default. Agents never get payment-card details. Agents that break the terms get blocked.

Two details matter more to brands than the headlines.

First, Walmart says the Gateway will help agents find a customer's "usuals," like "a particular brand of milk," without handing the agent the shopper's purchase history. Second, Walmart says "the prices that an AI agent sees are the same prices that a human would see on Walmart.com."

Why is Amazon keeping agents out?

Amazon's position is consistent, if you look at the business model. Muse users trying to shop Amazon now get a pop-up saying "continued access by an unauthorized AI agent violates Amazon's Conditions of Use." According to GeekWire, Amazon says Meta never told it Muse would access the store, that the agent doesn't identify itself when it browses, and that it appears to capture and store customer credentials. Amazon also sued Perplexity over its shopping agent earlier.

Amazon spokesperson Maxine Tagay told The Verge that third-party apps buying on behalf of customers "should operate openly and respect service provider decisions about whether or not to participate."

GlobalData Retail's Neil Saunders gave Axios the less diplomatic version: Amazon is protecting "its closed ecosystem" and likely fears agents could undermine its advertising business. That's my read too. An agent that compares prices across five retailers doesn't see Sponsored Products.

Amazon isn't anti-agent. It wants the agent to be its own. Alexa for Shopping, which folded Rufus into Alexa+, already offers price alerts and automated buying inside Amazon.

What does this split mean for CPG brands?

The practical effect showed up in The Verge's testing. Asked for the best per-piece price on granola bars, Muse came back with Sam's Club, Walmart and Costco, then admitted Amazon might have been cheaper but it was blocked from checking.

Think about that from the brand side. If your granola bar is priced sharply on Amazon and nowhere else, an outside agent can't see it. The shopper gets a recommendation for whoever is visible.

So the agent world is splitting into two lanes.

The open lane: Walmart, Sam's Club and Shopify stores. Shopify and Meta announced Shop Pay checkout for Muse across Shopify stores on September 21. Agents in this lane read structured catalog data. They don't look at your lifestyle photography, and they can't be swayed by a clever hero image. They compare attributes, price, pack size, availability and reviews.

The walled lane: Amazon, where the agent is Amazon's own assistant reading Amazon's own catalog, reviews and Q&A. We covered how to work with that in our Alexa for Shopping playbook.

Here's the opinion part. For CPG, the "usuals" detail in Walmart's announcement is the sleeper. Most of what CPG sells is replenishment: detergent, coffee, diapers, pet food. If agents get good at quietly reordering a household's usual items, the first purchase through an agent starts to look a lot like the first purchase in a subscription. Getting picked once may count for more than it used to.

Axios also notes Walmart's mix leans toward practical items like groceries and home goods, the kind of things shoppers are more willing to let AI recommend, according to University of Virginia professor Luca Cian.

Is any of this happening at real volume yet?

Not much, and it's worth being honest about that. Axios cites Coveo's 2026 Commerce Relevance Report: only 16% of shoppers are currently comfortable letting an AI assistant find and buy products for them. The Verge's hands-on test found Muse blocked by Target's bot check one day and let through the next.

Agent Profiles aren't live yet, and the Gateway is a beta.

But there are signals the curve is bending. Walmart U.S. CEO David Guggina said in September that Walmart's own assistant, Sparky, grew 60% quarter over quarter, per Axios.

My take: you don't need an "agent strategy" deck. You need clean data before the volume shows up, because fixing catalog data takes months and the holiday season isn't the time to start.

What should CPG brands fix first?

This is the short list I'd work through, in order.

Walmart item attributes. Fill every attribute Walmart's category template offers: size, count, flavor, dietary claims, ingredients. An agent filtering for "gluten-free, under $1 per bar" can only match what's written down.
Pack size and unit price consistency. Agents compare per-unit prices across retailers. If your 12-count on Walmart and 10-count on your site confuse the math, you'll lose comparisons you should win.
Facts that match everywhere. Claims, ingredients and sizes should be identical on Walmart, Amazon, Target and your own site. Engines cross-check, and mismatches cost trust. This is the core argument in our unified marketplace data piece.
Stock accuracy. An agent shopping for a household's usuals will skip an item that's out of stock, and may not come back.
Your own site's agent rules. If you sell DTC, check that your bot settings don't quietly block the agents you'd actually welcome. Our robots.txt guide covers the crawler side.
Amazon listings for Amazon's assistant. The walled lane still needs work. Complete bullets, Q&A and product schema on your own pages all feed the answers.

How do you know whether agents can find your products?

You test it the way a shopper would. Ask ChatGPT, Gemini, Perplexity and Amazon's assistant the questions your buyers actually ask: "cheapest high-protein bar per serving," "unscented laundry pods for sensitive skin," "best dog food for a senior lab." Note which retailer and which product each one names, and whether the facts are right.

Do it every week for your top SKUs, because answers shift. Our guide on how to measure AI visibility lays out the prompt set and the metrics, and SearchPanel automates that tracking per SKU across engines.

The bigger point is that marketplaces and AI visibility have become one job. The listing work on Walmart and Amazon is now the AEO work. That's why XENA built XENA 360 around both: the team runs the marketplace accounts and watches how AI assistants describe the products, in one place. You can see published results on our customers page.

What happens next?

Watch three dates. Agent Profiles were promised within a month of October 21, which puts them live just before Black Friday on November 27. The Gateway beta will show which agents Walmart actually lets in. And watch Amazon: if outside agents start to carry real purchase volume, a hard block gets more expensive to keep.

Until then, assume two lanes. Make your products easy for any agent to read on the open side, and keep feeding Amazon's own assistant on the other.

Frequently asked questions

What is the Walmart Agent Gateway?

It's a connection Walmart is building so AI agents can search and buy from Walmart and Sam's Club through MCP (Model Context Protocol) instead of clicking through the website like a person. Walmart announced it on October 21, 2026 and said it will start as a beta for companies and developers.

What are Walmart Agent Profiles?

A separate profile inside a customer's Walmart account for each AI agent they authorize. Walmart says customers will see each agent's purchases separately, can require approval for every purchase or set a spending limit such as $50, $100 or $200, and can delete the profile at any time. Walmart said they launch within a month of the announcement.

Why did Amazon block Meta's Muse?

According to GeekWire, Amazon says Meta didn't tell it Muse would access its store, that the agent doesn't identify itself when browsing and that it appears to capture customer credentials. Muse users see a message saying continued access by an unauthorized AI agent violates Amazon's Conditions of Use.

Do AI agents see different prices than shoppers on Walmart?

Walmart says no: 'the prices that an AI agent sees are the same prices that a human would see on Walmart.com, with applicable offers and fees shown before purchase.'

Should a CPG brand do anything about agent shopping yet?

Yes, but it's mostly work you should already be doing: complete product attributes on Walmart and your own site, consistent facts across retailers, accurate stock, and tracking how AI assistants describe your products. Only 16% of shoppers say they're comfortable letting an AI assistant find and buy for them, per a Coveo report cited by Axios, so this is about being ready, not panic.

Next step

Find out which of your products AI recommends.

Score a product across ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews with a free SearchPanel audit. Or book a meeting and we'll walk through a live AI Visibility Score with you.

Run a Free SearchPanel Audit Book a Meeting

Sources

XENA results quoted in this article are as published by XENA Intelligence on the linked pages. Third-party facts are from the linked official documentation, checked Oct 28, 2026.