Agent shopping has been “around the corner” for a while. A year ago, OpenAI tried to launch instant checkouts within ChatGPT[1]. Only 12 of Shopify’s millions of merchants went live with it, and OpenAI scaled it back after six months.
Walmart reported that conversion rates for items purchased inside ChatGPT were three times lower than for items that sent shoppers to Walmart.com. While merchants valued it as a discovery tool, they found that users weren’t eager to checkout from within the chat interface itself[2] This was the moment when people realized that e-commerce was more complex than they originally thought.
On September 8, Meta released Muse, a personal AI agent for users that is able to do much more than a traditional chat app. Modeled after the popular personal agent OpenClaw, Muse is capable of proactively managing emails, travel planning, and shopping online. Last week, it surged to become the number one app on the App Store, passing ChatGPT [3], with over 5 million downloads in its first week.
Muse offered a different approach to shopping - unlike ChatGPT, which could do research and surface the final checkout step to you, Muse could take full control of the shopping process and even handle your card information, making both decisions and payments on your behalf.
OpenAI recently followed up with Dots[4], its own always-on agents that can work on your behalf. Another firm, Instinct, recently raised a billion dollars in funding. Instinct allows users to interact with its proactive agent through text or calls, and give it tasks such as booking travel and handling subscriptions[5].
Muse sparked a divided reaction from online retailers - while Amazon swiftly blocked it, Shopify enabled eligible merchants to sell through Meta surfaces, including Muse, with direct checkout active by default for eligible stores[6] More on why below.
Muse doesn't only shop when prompted. It can set reminders to check pages daily for updates and let users know when a deal they might like comes up. As a merchant, you have to rethink holiday sales, pre-orders and coupon codes to fit this new paradigm.
This also opens up ways to offer more specialized deals, with the confidence that users will be able to find and keep track of them, representing a significant shift in the way commerce is done on the internet.
Agents also disrupt time-sensitive sales. Muse touts being able to grab concert tickets for a user's favorite artist as soon as they drop. It is promoted as being able to enter queues and act like a human, potentially bypassing anti-scalping measures. For platforms which want to stop this, it is important to be able to detect agents and enforce policies against them (more on this later)
There are many possible reasons for Amazon to have blocked Muse from its platform. Unlike Shopify, which serves millions of merchants through their own storefronts and channels on platforms like Google and Instagram, Amazon keeps shopping on its own website and app. Retaining users on its platform can thus become important to the company, because it allows them to leverage ad revenue and user analytics. They also keep control over the user experience, allowing them to surface suggested products and promote Prime benefits in the checkout flows.
Because a shopping agent can bypass shoppers’ interactions with Amazon’s ads, analytics, and checkout flow, it’s easy to see why Amazon might want to limit access.
Because Shopify has more flexible sales channels and does not rely on a single storefront as its core business model, it can integrate with Muse to increase merchant visibility and customer activity without threatening that model.
By default, Muse will use a browser in its VM to browse through product pages, add items to cart, and check out. Meta uses a secure VM for Muse, storing sensitive info in a Credential Store to keep it separated from the agent. When Muse launched, Meta said it initially used Link by Stripe for checkout. Link's agent wallet generates a one-time-use card, so Muse doesn't receive the user's actual card details.
This can take several minutes, and is inefficient for an agent that could directly interact with the platform’s API instead. To speed up shopping for agents, Shopify introduced Shopify Catalog, which enables programmatic access to a store’s products, variants, and cart functions. With this, agents can navigate product categories, choose products and variants that suit the request, as well as add them to a cart. This support is enabled through the Universal Commerce Protocol, or UCP for short.
Agent browsing vs. Agent UCP: Shopping with UCP gives agents direct access to the product catalog and cart, instead of having to navigate through a web UI.
Shopify also supports storefront features that make it easier for agents to navigate and interact with your store, such as an /agents.md with helpful resource endpoints and a sitemap, or a `/.well-known/ucp` endpoint describing the Universal Commerce Protocol support.
But UCP-compatible platforms still need a payment handler to complete checkout; Shopify’s Shop Pay handler supports delegated payments through Shop Pay tokens. To let Muse fully take advantage of its autonomous shopping ability, Shopify has added a Shop Pay payment handler specifically for agentic/UCP checkouts. Shop Pay is Shopify’s payment solution, which securely stores buyers’ payment and shipping information to streamline checkout. It accomplishes this by storing buyer payment and shipping information[7]. By expanding it to support agents, Shopify allows Muse to use this saved information to complete checkouts, without revealing the user’s card information to the agent. This is an important part of the integration.
If you are a merchant, you are able to see orders processed through integrated AI channels like Meta Muse show up in your Shopify dashboard, as well as a new “Agentic” tab showing AI browsing analytics. If you wish to disable direct checkout on Meta, you can do so in your Shopify admin; customers can still discover your products on Meta and be redirected to your online store to complete their purchase. For more information on selling through Meta, check out Shopify Help: Selling on Meta.
While the Shopify dashboard shows sales completed through AI channels such as Muse and products browsed, it leaves a lot to be desired in the area of analytics. You can’t see exactly which agents are arriving, how they navigate through your website before reaching the products they are looking for, and whose behalf they’re shopping on.
Whether you’ve already enabled agentic rails on your store or are still on the fence, it can be useful to get a picture of traffic on your website by checking what agents are already browsing.
This is more difficult than it looks on the surface, because unlike crawler and user bots operated by services like ChatGPT and Claude, agents like Muse which browse in a VM will not identify themselves as such. Also, any bot can spoof its request headers to evade detection. If a ticket scalping agent wanted to fly under the radar, it could easily spoof its User-Agent and pretend to be a human. For more details, read our article on agentic vs bot traffic on the internet.
This is why we at Vouched created a platform that combines multiple methods and heuristics to accurately spot agent traffic on the internet. We created three methods: a marketing pixel, a client-side SDK, and a WASM engine that can run on the edge or as middleware on your server to analyze request attributes and classify traffic. The platform is called Agent Checkpoint, and it offers an easy way to see the volume of AI traffic on your website.
With this, you can answer crucial questions such as: Which agents are visiting my website? Which products are they browsing? Are they being led to them from any of my content pieces or blog posts? How long are agents sticking around and browsing? This helps you have a complete picture of your traffic to decide how to handle AI sessions.
Agent Checkpoint doesn’t just offer detection: the Enforce feature allows you to describe product access policies in natural language (such as “Block all requests to /categories/office-supplies from ChatGPT”) and turn it into a Cedar policy that runs on every request. This gives you fine-grained control over what agents can access and what they can’t on your platform.
▶️ To get started with Agent Checkpoint detection, visit our documentation.
Additionally, if payment authorization alone is insufficient for you, and you need Know-Your-Agent capabilities, KYA-OS is an open source standard that fills that gap. Designed to easily work on top of existing standards like UCP, A2P, and x402, it adds the missing identity layer that enables advanced agentic use cases such as multi-agent delegations, identity anchored through web credentials, ledgers, or personal keys, and revocation mechanisms. To read more, check out https://www.kya-os.org/.
Meta’s Muse is closing the gap in AI agents’ end-to-end shopping capabilities. It’s already at No. 1 in the App Store, and other agents are expected to follow soon. We anticipate a sharp rise in AI-driven product discovery and checkout in the coming weeks. Making sure your store is ready to handle and serve this traffic is crucial!
Sources:
https://rye.com/blog/openai-chatgpt-checkout-agentic-commerce
https://www.modernretail.co/technology/what-went-wrong-with-chatgpts-instant-checkout/
https://tech.yahoo.com/ai/meta-ai/articles/metas-muse-becomes-app-stores-172320576.html
https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts