AI agents: opening or closing your store has become a leadership decision

Section 1 · The news

BigCommerce opens its stores to AI agents, without asking merchants first

On October 21, 2026, Commerce (the Nasdaq-listed company behind BigCommerce) and Cloudflare announced WebMCP for BigCommerce stores. In practice, an AI agent running in a shopper's browser can now search products on the site, compare them and add them to the cart on the shopper's behalf.

The press release gives a simple example. A shopper asks their agent for a lightweight jacket in their size, under $150. The agent finds a product on a BigCommerce store, adds it to the cart and hands the shopper a link to checkout on the merchant's site.

The line that matters for an e-commerce director is this one: the feature is “turned on by default for eligible BigCommerce merchants.” No code to write, no change to the site. Merchants who do not want it can turn it off in a new settings menu, which also manages all of Commerce's AI connections.

What stays with the merchant According to the release, prices and stock come straight from the store, and “checkout, the customer relationship and the brand stay with the merchant.” The agent prepares the purchase. The shopper still pays on your site.

Section 2 · Explainer

MCP and WebMCP hand agents a user manual for your store

An AI agent that shops has a problem: e-commerce sites are built for humans. It has to guess where the search bar is, which button adds to cart, which size is in stock. It reads the page like a picture, often gets it wrong, and the site may take it for a malicious bot.

MCP (Model Context Protocol) is an open standard, introduced by Anthropic in November 2024, for connecting AI assistants to the systems where data lives. Instead of guessing, the agent receives a list of possible actions, clearly described: “search for a product”, “check stock”, “add to cart”. WebMCP brings that idea to the web page itself: the site tells the browser agent directly what it sells and what it can do.

Sharon Gee, who leads AI product at Commerce, sums up the intent: WebMCP “replaces guessing with a handshake.” The site and the agent agree on what is possible, instead of the agent feeling its way around.

Section 3 · The stances

Amazon closes, BigCommerce opens, Walmart picks its partners

The same week, Amazon showed the opposite stance. It blocked Meta's Muse agent, launched in September 2026 and able to shop for its user. In a statement quoted by Cybernews, Amazon said that “continued access by an unauthorized AI agent violates Amazon's Conditions of Use, to which our customers have agreed.” In mid-October, users also reported that Amazon was stopping agents from Grok, xAI's assistant, from placing orders. Amazon has not confirmed that second block.

Walmart took a third path. On September 24, 2026, Meta announced it was adding Walmart to Muse, alongside Best Buy, Sephora, Ulta, Wayfair and American Eagle Outfitters. At the same time, Walmart is building its own assistant, Sparky, in its app since June 2025, which helps shoppers find products and sums up reviews.

Why Amazon can afford to close Amazon is a destination in itself: its customers come to it directly. It does not need other companies' agents to be found. An average brand does not have that leverage: if it closes, the shopper's agent buys somewhere else.

Section 4 · The impact

Each stance is paid for in conversion, promotions, data or measurement

Nobody knows yet how much agents will sell. None of the three companies has published sales figures for purchases made through agents. But we can already describe what each stance shifts on four topics an e-commerce director tracks every day.

The most underestimated topic is measurement. An agent does not see your free-shipping banner, does not click “complete the look” and does not come back three times to the product page. If its traffic blends with human traffic without being flagged, your dashboards and A/B tests measure a mix nobody chose.

“The default setting is already a decision. The only question is who made it.”

Section 5 · The decision

The decision belongs to e-commerce leadership, with a name and criteria

On BigCommerce, the choice comes down to a switch in a settings menu. That is exactly the danger: the decision can be made by a developer, an agency or nobody at all. Yet it affects revenue, margins, data and the brand. Here are the questions to settle before leaving the switch one way or the other:

1. Who owns the decision? One named person in e-commerce leadership, with legal and data at the table. The tech team should not decide alone.
2. What is our stance today? Check what your platform has turned on by default, and what your robots.txt file and bot-protection tool let through.
3. Which agents matter to our customers? The ones they actually use, rather than the ones in the headlines.
4. What should the agent see? Prices, stock, promotions, bundles: anything not exposed as an action may be ignored.
5. How do we recognize agent traffic? Without tagging, you cannot measure what it brings or what it breaks.
6. What are the limits? Excluded products, maximum quantities, return conditions: what an agent must never do on your store.

Section 6 · The test

Open on a limited scope, and measure before rolling out

Between opening everything and closing everything, there is a more careful option: open part of the catalog or part of the traffic, with measurement planned in advance. It is the logic of an A/B test, applied to an access decision.

This test will not tell you what agents will sell in three years. It will tell you, on your own store, whether a cart filled by an agent turns into an order, at what order value, with what returns. That is enough to move from a setting you inherited to a decision you own, and to revisit it as the numbers come in.