Meta Put a Shopping Agent in WhatsApp. Your Product Page Is Now Part of the Conversation.

Meta launched Muse on September 8, an AI agent that can browse, fill forms, negotiate and make purchases from a conversation in WhatsApp.

That puts a new kind of buyer on your product page. It arrives with a task, a budget and a user behind it who doesn’t want to spend 20 minutes comparing tabs.

Meta says Muse can keep working after the person closes the app, then return when it needs approval for something sensitive, including a purchase. It can use a dedicated browser, connect to apps including Instagram and Facebook, and check out with Stripe Link. Muse’s product page says Shop Pay is planned next.

For a retail brand, the immediate implication is obvious. A buyer can hand the comparison work to an agent and judge the answer it brings back. The product page, inventory status, pricing, returns policy and support information become inputs to that answer.

For a B2B software company, the same thing happens a little earlier in the buying process. An AI agent doesn’t need a sales call to find the pricing page, compare implementation claims or notice that the security documentation is six months old. The salesperson gets the meeting after the research has already formed a shortlist. This is the same structural shift behind AI shopping traffic and branded search.

The visit is part of the ad system

Meta draws a useful boundary around Muse. It says conversations and data inside a person’s Muse virtual machine don’t go to its ad systems.

But Meta’s own safety documentation says browsing by Muse appears as the person’s activity. Ask it to shop a clothing site and that visit can affect the ads the person sees on Instagram. The agent doesn’t sell an ad against the chat. It can still create a behavioral signal once it leaves the chat.

That means product information needs two jobs at once. It needs to persuade a person who lands on the page, and it needs to let an agent finish a task without guessing.

The second job is less glamorous. Keep prices and availability current. Put shipping, returns, eligibility and product limits where a browser can find them. Make the comparison page say what you do without forcing a form fill. Treat vague claims as a sales problem, because an agent can’t call the founder to resolve them.

Meta is building guardrails around the actions that make this consequential. Tarek Sheasha, Software Engineer & VP at Meta Superintelligence Labs, wrote that the company had handed “our inboxes, our calendars, and a shell to a piece of software and let it run unattended,” and that it “didn’t always work out as planned.”

The caveat keeps the claim proportionate. Muse is US-only at launch. Meta hasn’t published usage, conversion or purchase-completion data. It also says the product will sometimes make mistakes, which matters more when the agent can act than when it only recommends.

The page is no longer the last step

Marketers have spent years treating the product page as the landing spot after the ad, the email or the search result did the hard work. An agent changes the sequence. The page can become part of the research process itself.

A good page gives the agent facts it can compare. A bad one leaves it to infer the terms, or sends the buyer to a competitor with cleaner answers. Product feeds already carry the same burden in AI commerce.

Meta hasn’t proven that people will hand meaningful shopping decisions to Muse. It has made the behavior possible inside one of the largest messaging surfaces in the US. Brands don’t need a separate “Muse strategy” this week. They need to stop treating the site as copy written only for a human who has time to hunt.

The practical test is simple. Pick one product or plan and ask an AI agent to compare it with two alternatives using only public information. The gaps it finds are the parts of your buying process a person has been papering over.