BeaconSpec

← All articles

Agentic commerce • Field briefing

Walmart built its own door for AI agents.

On October 21 Walmart announced the Agent Gateway, a server that approved AI shopping agents connect to so they can search its inventory and buy, under rules Walmart wrote. The same day, BCG reported that only 18% of retailers think their marketplaces are ready for agents. Walmart had the engineers to build its own door. Most businesses don't.

David Soden  •  Agentic commerce  •  October 26, 2026  •  7 min read
A person holds a blue key card against the electronic lock of a door while gripping the handle, with a second person looking on.
Walmart's gateway works like a key card: approved agents get in, and the rules come with the key.

The Walmart Agent Gateway is an MCP server, the open protocol AI assistants use to call outside tools, built for shopping agents Walmart has approved, TechCrunch reported. It already has ties to Meta's Muse, ChatGPT and Gemini. Walmart wrote terms for the agents that come through it: an agent can't act beyond what the shopper asked for, can't touch payment details, and gets blocked if it breaks the rules. Walmart's returns policy still applies, and Walmart says agents see the same prices people do. A second piece, Agent Profiles, rolls out in November and lets a shopper give each agent its own profile with a spending cap or an approval step on every purchase. Walmart hasn't given a launch date for the gateway or published how an agent gets approved.

TechCrunch's headline put the contrast plainly: while Amazon blocks AI agents, Walmart rolls out the welcome mat. Both companies faced the same question. Amazon's answer is to keep agents out. Walmart's is to let them in through one door, on its own terms.

Walmart decided that if agents were going to shop its shelves, they would come through a door it built, on rules it wrote.

What's happening

Agents are becoming a real audience, and every seller now has to decide how they get in.

The audience is no longer hypothetical. Meta told investors on its third-quarter call that Muse, its shopping-capable assistant, passed 10 million users in about six weeks, with shopping one of the top things people do with it. More than a quarter of engaged users who shopped through Muse spent over $200 on average, according to TechCrunch's report of the call. Mark Zuckerberg also teased a network where shoppers' agents talk to businesses' agents, without details.

The platforms are answering for their merchants. On October 21 Cloudflare said Commerce.com storefronts, almost 100,000 of them, now give agents tools to search products, build a cart and get a checkout link, with no code from the merchant and an opt-out in the dashboard. That runs on WebMCP, which is still an experimental browser standard. A store on a platform like that gets an agent path whether it asked for one or not. A business on its own systems has nobody to switch it on.

A shopper in a white sweater holds a red shopping basket in one hand and checks a smartphone with the other, beside wooden shelves of goods.
More shoppers are handing part of the job to an agent. Where that agent goes depends on who it can talk to.

What's changed since our last briefing

Last week we looked at who made it into Google's cart. This week the biggest retailer on that list showed what its side of the door looks like.

Our last briefing covered Google's Universal Cart launch and its roster of big retailers and Shopify brands, each with engineers or a platform behind it. Walmart was on that list. Its gateway shows what a large retailer builds once it decides agents are customers: one interface, approved callers, written limits, and its own prices and policies.

BCG put numbers on how far everyone else is from that. Its study of agentic commerce, published the same day and built on a survey of 50 of the 300 largest global retailers and 100 leading brands, found only 18% of retailers say their marketplaces are ready for agents. The obstacles they named were fragmented legacy systems, weak product data and missing integration layers. BCG also asked how retailers treat outside agents today.

How retailers treat outside AI shopping agents

Share of retailers in BCG's survey, by their stance on agents they don't run themselves.

Let any agent read data 42% Allow only chosen agents 32% Still deciding 20% Block all agents 6%

Source: BCG, "How Retailers Can Reinvent the Digital Flywheel for Agentic Commerce," October 21, 2026.

Almost three in four retailers have already said yes to agents in some form. Saying yes is the easy part. BCG's 18% figure is about whether the systems behind that yes can answer an agent, and most can't yet. The same study found 87% of retailers never share customer data with agents, and only 4% of consumers would let an agent finish a purchase on its own. Walmart's spending caps and per-purchase approvals are aimed squarely at that 4%.

Why this matters to you

If you don't publish your own door, agents reach you on someone else's terms, or they don't reach you.

BCG estimates that more than half of ecommerce spending could be agent-assisted by 2030. That's a projection, but it comes with a warning that matters now. BCG expects demand routed through agents to carry a fee, and points to online travel agencies, which take 15% to 25% of each booking, as the model. And 91% of the retailers it surveyed named losing the customer relationship as their top worry.

Shoppers already lean toward the big intermediaries when they send an agent. In a PYMNTS Intelligence survey of 2,191 US consumers, published October 23, 40.2% said that with price and shipping equal they'd have an agent buy from a large marketplace, against 16.1% who'd pick the brand's own site. Hotels show where that leads. Hospitality Net reported that Google's AI Mode now completes bookings for ten large chains and online travel agencies, and on October 1 the rule that kept independent hotels' free listings beside OTA results ended.

A dark server room lit in blue, with tangled network cables hanging from racks above a single computer monitor.
Retailers told BCG the blocker is the systems they already run, which were never built for an agent to call.

Put those together and the choice is plain. An agent will buy from you through an interface you publish, through a marketplace or platform that sets the rules and may charge for the privilege, or not at all. A website by itself doesn't pull agent traffic. What earns you a place is an interface agents can call, published to the standards they already use.

Why we're built for this

Walmart built its door by hand. We build yours from the APIs you already run.

BeaconSpec is the discovery layer for agentic commerce. You already run APIs behind your catalog, booking engine or ordering system: REST, OpenAPI or GraphQL. We turn them into one curated MCP server, the same kind of interface Walmart chose, and publish its UCP profile at your own /.well-known/ucp, so agents can find it and call it.

The rules are yours. You pick exactly which operations an agent may use, and anything you leave off doesn't exist as far as the agent is concerned. When a task needs the customer signed in, like checking an order, we handle the standard OAuth login so the agent works inside that customer's own account, with no shared passwords. The prices and stock an agent sees come straight from your systems, as they are today.

We host it, or hand you a sealed container to run in your own environment. BeaconSpec isn't a platform, a marketplace or a payment network, so the door belongs to you and no one else's terms decide who comes through it. Publishing an interface doesn't send shoppers to you on its own. It makes you something an agent can actually call when a shopper sends one your way. The standards keep changing, and keeping that interface correct as they do is our job.

The one thing to remember

Walmart built its own door for AI agents, with its own rules. Most retailers can't, and BCG says their legacy systems are the reason.

The alternative is reaching agents through a marketplace or platform that writes the terms and, BCG expects, will charge for the traffic.

Your APIs already hold the answers an agent needs. The door is the part that's missing.

Worth watching over the next month: how Walmart approves agents for its gateway and which ones get in; Agent Profiles going live in November; whether Meta says more about a network of business agents for Muse; and whether Google adds merchants off the big platforms to Universal Cart.

BeaconSpec exists for exactly this shift: making your business discoverable and transactable by AI agents over UCP, on an interface you publish and control, kept correct as the standards underneath it keep changing.

See what we do  •  Read more articles


David Soden writes about automation, web strategy, and building durable technical systems for businesses. Photography via Pexels (Ketut Subiyanto, Kampus Production, panumas nikhomkhai). Figures cited are drawn from TechCrunch, Adweek and Chain Store Age on the Walmart Agent Gateway and Agent Profiles (October 21 and 22, 2026); BCG, "How Retailers Can Reinvent the Digital Flywheel for Agentic Commerce" (October 21, 2026); Cloudflare's blog post on WebMCP for Commerce.com storefronts (October 21, 2026); TechCrunch on Meta's third-quarter earnings call and Muse (October 21, 2026); PYMNTS Intelligence on where shoppers send AI agents (October 23, 2026); and Hospitality Net on Google AI Mode hotel bookings (October 19, 2026).