Meta's Muse agent now finds products and pays for them on its own, Shopify has opened its checkout to agents like it, and the brands that get bought from will be the ones whose data an agent can read.
On 8 September, Meta launched Muse, a consumer AI agent that browses and checks out by itself. When independent developers looked inside its code, they found a checkout route built specifically for Shopify stores. On 28 September, Shopify opened its checkout to AI agents.
For the past few years, agentic shopping has been a prediction. A consumer tells a piece of software what they want, and the software finds it, compares the options, pays and arranges delivery, without the consumer ever landing on a brand's website. The hard part was always the infrastructure. An agent has to read a catalogue, understand stock and sizing, get through a checkout and pay with credentials the merchant will accept. In September, Meta and Shopify put the missing pieces in place, and a consumer in the US can now hand a purchase to an agent and never see the product page.
Muse is available in the US only for now, on mobile and the web, and it can use a browser and complete a purchase on someone's behalf. Each purchase runs on a one-time card generated through Stripe Link, so the merchant never sees the shopper's real card details.
How Muse pays depends on where it is shopping. Independent analysis of its code found three separate checkout paths. On an ordinary website, it types card details into the merchant's form, much as a person would. On a Shopify store, it goes through Shopify's own systems instead, apparently using Shop Pay over the Universal Commerce Protocol, a shared standard that lets agents find products and check out. Meta's own announcement names neither protocol, so that detail is the developers' reading rather than Meta's confirmation. A third path handles sites that already use Stripe.
When Meta showed Muse at its Connect event on 23 September, it named Best Buy, Gap, Sephora, Walmart and Wayfair as retail partners. It also added Shop Pay and PayPal as payment options and said Muse could reach Shopify's entire product catalogue. Muse went on to top the US and Canadian app charts, ahead of ChatGPT.
Tim Richardson, our CEO, sees Muse as the first fully fledged consumer AI capable of disrupting the traditional end-to-end shopping journey. In his words, it is "a full on autonomous agent that can do consumption on one's behalf."
Shopify's side of the deal arrived on 28 September, roughly a week after its partnership with Muse was announced. Shopify has opened its checkout, Shop Pay included, to agents working inside a shopper's browser. An agent can now inspect a checkout and change details such as the delivery address or delivery option. Once the buyer has authorised it, the agent can place the order. None of this relies on the agent taking screenshots or scraping the page, and it is rolling out to all eligible Shopify merchants.
Muse is not the only agent using it. On the same day, Instinct, an assistant people text or call to get things done, launched its own Shopify integration and announced a $1bn raise at a $10bn valuation. A month earlier it had been valued at $2.5bn.
The most useful thing about September is the direction it points in. The crude way for an agent to shop is to act like a person: loading a web page, reading what is on screen and filling in forms. That approach is slow and fragile, and merchants have little visibility of who is actually buying. What Shopify has built, and what Muse uses on Shopify stores, works differently. The shop hands the agent clean data and a checkout over a defined connection, and the agent acts with the buyer's authorisation.
The payments industry is heading the same way. On 10 September, Ant International, Mastercard and Visa said they would work together on a shared Know Your Agent framework, setting common principles for identifying and onboarding AI agents across their networks so that a trusted shopping agent can be recognised wherever it tries to pay. There is no technical specification and no launch date yet, so this is a direction of travel rather than something brands can act on today. But it answers the question merchants have been asking about agentic shopping from the start: how does a store know the agent in front of it is legitimate?
Put the two together and the shape of agentic commerce is becoming clearer. Standards like the Universal Commerce Protocol give agents something proper to read. Know Your Agent work gives merchants a way to trust who is reading it. That is engineering work: product catalogues, structured data and a checkout an agent can actually get through.
This is close to the sequence we have been describing for the past year. In October 2025, when OpenAI launched Instant Checkout inside ChatGPT, Tim wrote on Offcuts about three stages of what he calls external AI: AI search, where products are discovered inside a chat interface; AI conversion, where the purchase happens inside that interface; and agentic shopping, where the consumer hands the decision and the purchase to an agent. At the time, agentic shopping was a future state, with order management and returns still to be solved.
By March 2026, the Offcuts piece "The .com Is Not Dead. It Is Getting Smarter." had turned that into a question of foundations. It argued that clean catalogues and connected systems were becoming the new baseline, and that brands should audit their product data for agent-readiness before any of this reached consumers at scale.
Our AI services guide has carried the same warning: "AI agents are beginning to shop on behalf of consumers without them visiting a brand website." It also makes the distinction that matters most now: "Being recommended is not the same as being transactable." We have been consulting and writing on AI in commerce since 2023. What changed in September is that the third stage now has a consumer product with real retail partners behind it, and Shopify has opened its checkout to it.
AI-driven shopping is still smaller than search, but it is growing fast. Similarweb's State of Ecommerce 2026 report, published on 10 September, found that direct referrals from AI platforms grew by more than 200% over the past year. It also found that 89% of the time, consumers who use AI in their shopping research also use search, so AI is adding to search rather than replacing it. And when an AI chat recommends a brand, Similarweb puts that brand's advantage over its competitors at two to one.
So the first job is being the brand an agent names. The second, now that agents can buy, is being a brand an agent can actually complete a purchase with. Both come down to product data. A product page does a lot of work for a person: photography, copy, reviews and a size guide all help them decide. An agent reads data instead. If sizing, materials, stock levels or the returns policy are missing or buried in an image, the agent will either get the answer wrong or move on to a competitor whose data answers the question.
Checkout is the next pressure point. Many Shopify stores run upsell apps and custom fields that were designed for a person clicking through. A brand whose checkout has been heavily customised will find out quickly whether those additions still work when an agent is filling in the form. Bot protection needs the same review. Rules written to keep scrapers out can just as easily shut out a legitimate buying agent acting for a real customer, and a brand will not see the sale it lost.
For Shopify brands, getting agent-ready means structured product data and readiness for the Universal Commerce Protocol. For brands on Centra, there is more to do. We have not yet seen Centra make a move equivalent to Shopify's, and feed and catalogue architecture on Centra is usually spread across several brands and markets. That gives Centra brands time to get their product data in order before a similar route opens, and it is where we would start.
Muse is US only, and AI platforms still send far less traffic to stores than search does. The Know Your Agent framework has no specification yet. None of this means every brand will see agents buying from it next month.
The bigger questions are ones no platform will answer for a brand. Which agents should be allowed to buy? How much product and customer data should they see? How far should a store go in stripping back a checkout experience it has spent years building so that an agent can get through it? Those are commercial and brand decisions, and they need someone who knows the business to make them. The technical work of making a store agent-ready is only worth doing once those calls have been made.
The person still decides what they want. Increasingly, an agent will do the finding and the buying. The brands those agents buy from will be the ones whose product data answers the agent's questions and whose checkout an agent can complete. For Shopify merchants, that became a practical requirement in September. For everyone else, it is a matter of when.
