On 8 September 2026 Meta launched Muse, a personal AI agent. Where Meta AI answers questions inside Meta's apps, Muse is built to take actions: write and send emails, make bookings, fill in forms and buy things, with payments through Stripe. It runs on Muse Spark, Meta's own model, and works in its own app on iPhone and Android, on muse.ai and inside WhatsApp. Meta added to it at its Connect event on 23 September.
What it does
- Tasks across the web: email, bookings, forms and purchases.
- Connectors to shopping services and other tools, which Meta expanded at Connect.
- Its own Muse email address, and on a Mac the ability to use the computer, both announced at Connect.
- A real-time avatar built on Meta's voice model, a small device called Muse Charm, and Muse on Meta's glasses in the coming months, also announced at Connect.
Privacy, as Meta describes it
Meta says Muse runs tasks in what it calls a Muse Secure VM, an isolated virtual machine, and that a Confidential VM, where the user holds the encryption key, is due later in 2026. It also says Muse data is not used by its ad systems. Those are Meta's claims; they matter for any client whose staff might one day use Muse with work accounts.
Who can use it, and what it costs
- The US at launch, for people aged 18 and over. Canada followed on 18 September. Some reports mention Mexico, which we could not confirm.
- A free tier, a Power plan at 20 dollars a month and a Maximum plan at 100 dollars a month, according to TechCrunch. Meta's help centre gives Power 500 million Muse tokens a week and Maximum 3 billion.
- It is not available in the UAE or Saudi Arabia.
How it has gone so far
Muse reached number one on the US App Store. Meta has not published download figures; app analytics firms estimated anywhere from about 900,000 downloads in six days to 2.8 million installs in twelve, depending on the firm and method, and TechCrunch reported it was outpacing ChatGPT's early mobile launch. Market reports put Meta's share price up more than 11 percent on 21 September.
There has been friction too. Amazon blocked Muse from using its store within two weeks of launch. And Nat Friedman, head of product at Meta's AI lab, acknowledged that Muse was heavily inspired by OpenClaw, an open-source agent.
Why agencies should care before it arrives
An agent that shops and books changes who, or what, a brand is talking to. Today a customer sees an ad, visits a website and decides. With an agent, the customer may ask it to find a table for Thursday or reorder a product, and the agent reads the web and decides what to suggest. Meta has not said how Muse chooses between businesses, and says Muse data is not used by its ad systems. What a brand can control is what an agent finds when it looks:
- Websites that load quickly, state products and prices plainly, and do not hide the checkout behind pop-ups.
- Booking and ordering that can be finished online, not only by phone or by WhatsApp message.
- Accurate business details on Google, on maps and on review sites.
- Product catalogues with clear names, prices and stock.
None of that is new advice, but an agent is less forgiving than a person. A site a human can muddle through may simply fail for an agent, and the agent will move on to a competitor. Gulf businesses that take bookings and orders mainly through WhatsApp conversations are the most exposed.
What to do now
- Explain to clients what Muse is, and that it is not in the Gulf yet, before they hear exaggerated versions.
- Test clients' booking and checkout flows as an impatient visitor would: can a form be finished start to finish, and are prices and availability on the page?
- Watch the Amazon block. Whether large retailers let agents in will decide how useful Muse becomes.
- Do not put clients' business accounts on US-only services through workarounds.
The free website traffic checker on this site gives a quick read on any site's monthly visits, a starting point for deciding which client sites to prepare first.
