Meta's Muse can send emails and make payments for you - what it means for business automation
Meta's Muse sends emails, books travel and makes payments for one person. Why that makes it the wrong tool for a business, and what to take from it anyway.
The news
Meta introduced Muse on September 8, a personal AI agent that does things rather than just answering questions. According to Meta's announcement, Muse sends emails, books travel, fills out forms, negotiates on your behalf, and completes online purchases through a Stripe wallet. You talk to it through an app or through WhatsApp. It keeps working after you close the app and comes back when it needs a decision. Meta says it asks for approval before it sends an email or spends money.
It runs in the US only, on iOS, Android and the web, with support for Meta's AI glasses promised later. Pricing is a free tier plus paid plans at $20 and $100 a month depending on how much you use it. The free tier reportedly covers up to 100 million tokens a week. Muse runs on a dedicated virtual machine Meta calls Muse Secure VM, with a separate "Sentinel" agent that vets every request to reach the internet, plus an audit log of everything it did.
What is actually new
A general-purpose agent that drives a browser, fills forms and holds a plan across hours is not new by itself. OpenAI, Google and Anthropic all ship a version. What Meta added is packaging: a consumer product, a phone app, a WhatsApp thread, a price most people will read as free, and a memory that carries your preferences between tasks so you stop re-explaining them. The pitch is that a normal person can hand off a multi-step errand and get it back done.
That is a real step. The interface is the product here, not the model.
What it means for a business owner
The obvious question is whether you can point this at your operations. Mostly no, and the reasons are worth understanding because they apply to every consumer agent, not just this one.
Muse is built around one person's context: your inbox, your calendar, your card. A business workflow runs on shared systems, on defined inputs and outputs, and on a result that has to be right every time rather than most of the time. "Book me a flight" tolerates a wrong guess because you are watching. "Reconcile these invoices against the bank feed" does not, and nobody is watching the 200th run.
The approval step that makes Muse safe for personal use also makes it impractical at volume. If it stops for a human decision before every email and every payment, it is not automating the work. It is putting a chat interface on top of it. Automation earns its keep when it runs unattended. An agent that needs you in the loop on each action is a slower way to do the task by hand.
Then there is the data question. Muse wants access to email, calendar, payments, health apps, shopping and smart-home devices. Meta says conversations do not feed its ad systems and that the agent cannot see your passwords or card numbers. For a personal account that is a judgment call you make for yourself. For a business holding customer data, wiring your operations into a Meta consumer product is a decision your contracts and your clients may have opinions about.
The honest caveat
This is not a knock on the technology. The same browser-driving, plan-holding capability inside Muse is genuinely useful in a business when it is scoped to one workflow, fed defined data, and wrapped in logging and error handling on infrastructure you control. That is the work we do. Muse is that capability in a consumer skin, and the skin is what makes it wrong for operations.
Miranda Bogen at the Center for Democracy and Technology put the risk plainly. When an agent gets something wrong, she warned, it can "damage someone's reputation, leak sensitive medical information, or even deplete bank accounts." That holds for the personal version. It holds harder for anything touching a business.
What to do about it
If you have been waiting for an agent you can run your business on, this is not it, and "just use Muse" is not a plan. Treat it as a signal instead. The parts that are genuinely hard, like operating web tools that have no API, holding a task across steps, and recovering when a page changes mid-run, are getting reliable enough that Meta will now sell them to consumers. Pick one workflow you want automated. Write down its exact inputs, outputs and failure cases. That document is the thing worth building on when you are ready.
Want this kind of system in your business? Book a free scoping call.