Meta's Muse Shops and Negotiates on WhatsApp for Buyers

Meta has introduced Muse, an AI agent that users control through WhatsApp and that carries out tasks on the web by itself, as reported by The Decoder. It opens a browser, fills in forms, negotiates on the user's behalf, and completes purchases through Stripe's Link once the user approves. If people order from you through WhatsApp or your website, some of the messages and clicks you get may soon come from software acting for a customer rather than the customer.

What Meta announced

Muse takes instructions in WhatsApp and works on its own machine in the cloud. According to Meta, it can send an email, book a trip, plan multi-step jobs, and keep running while the app is closed, checking back when it needs approval or when something changes. Meta gives examples such as selling a car for more, lowering a bill, or turning a saved Instagram recipe into a shopping list.

Payments run through Link, the Stripe service. Meta says Muse is the first AI agent covered by Link's purchase protection, which is meant to cover damaged or lost items, price drops and returns. Each payment uses a one-time card, so the buyer's real card details are not exposed. Shop Pay and a 1Password connection are planned, so the agent can use existing logins.

On security, Meta describes a walled-off virtual machine holding the credentials for connected services, plus a second agent called Sentinel that has to clear anything before it reaches the internet. Meta says Muse itself sees neither passwords nor payment methods, asks before sensitive actions, and shows a record of every step. Users choose which apps it can touch and how far the access goes. Meta publishes no numbers on how robust this is, and researchers have already shown that agent systems can be hijacked through manipulated content — with Perplexity's Comet browser, a doctored calendar invite was enough to take over a password manager account.

Muse starts in the United States on iOS and Android, with a free usage allowance and paid subscriptions above it. A link to Meta's AI glasses is planned.

Why it matters for a small business

Three things change on your side of the counter.

First, an agent reads your site the way a machine does. It looks for a price, a stock status, a size chart, a delivery time and a form it can complete. If that information lives in a photo, a PDF, or only in the head of the person answering your WhatsApp, an agent will either skip you or guess wrong.

Second, negotiation becomes routine. Meta explicitly says Muse haggles on the user's behalf — lowering a bill, pushing for a better price. A person negotiates once and gets tired. An agent will ask every time, in every channel, and it never feels awkward doing it.

Third, checkout gets stricter. A one-time card and purchase protection mean disputes over price drops, returns and damaged goods have a formal path. Your return policy and your published prices stop being soft statements and start being the terms you are held to.

None of this depends on Muse specifically. It is one product, launched in one market, and Meta's claims about it have not been independently measured. But the direction is clear enough to prepare for, because it costs little to prepare.

What to do about it

• Put the facts in text on the page: price, currency, stock, delivery window, return terms, sizes and compatibility. Machine-readable structured data on product pages helps more than a nicer product photo.
• Decide your discount policy before an agent asks. Write down what you will give and to whom — volume, prepayment, slow-moving stock — and give your team one script for both humans and bots.
• Check that your WhatsApp and web forms can be completed by a patient stranger with no context. Fewer required fields, clear error messages, no steps that only work if someone calls you.
• Make your published price the real price. Hidden fees that appear at checkout are exactly what an automated buyer will flag, dispute or abandon.
• Tighten approval on your own side. If your staff use AI assistants with access to email or accounts, define what those assistants may read and what they may send, and keep a log.

The practical takeaway is unglamorous. The businesses that handle agent traffic well are the ones whose catalog data is accurate, whose prices are stated plainly, and whose ordering flow does not rely on a phone call to finish. That work pays off with human customers too, which is a good reason to start now rather than wait to see whether Muse gets traction.