Issue #32: Meta just launched an AI called 'MUSE' that doesn't just answer questions, it does the work

Muse sends the email. Books the table. Fills out the form. Compares prices and checks out. All while your phone is in your pocket.

It's not a feature list from a sci-fi movie.

That's Meta Muse: launched September 8, 2026, for US adults, free to download, 2.5 million installs in two weeks.

The fastest consumer AI download since ChatGPT itself.

For three years, AI competed on who could generate the best answer.

Muse competes on something different: how much real work it can take off your plate, end to end, without you watching every step.

Muse is Meta's personal AI agent. Not a chatbot. Not an assistant that tells you how to do things. An agent that does them.

The technical difference matters here. Every AI product you've used until now: ChatGPT, Claude, Gemini - operates as a response machine.

You ask, it answers. You act on the answer.

Muse operates differently: you give it a goal, it figures out the steps, uses your connected apps and a dedicated cloud computer to execute them, and reports back when it's done.

It keeps working even after you close the app. Basically an AI agent but since Meta has rolled it out, much more massy.

Here's what it can actually do:

Email and calendar: connect your Gmail or Outlook and Muse reads, drafts, and sends emails on your behalf. It can manage your calendar: schedule meetings, block focus time, respond to invites. Tell it "clear my Thursday afternoon and reschedule anything that conflicts" and it does it.

Shopping and price comparison: give Muse a product to find and a budget. It browses, compares prices across retailers, applies promo codes, and checks out with your saved payment method with your approval before any transaction completes.

Travel booking: tell it your destination, dates, and preferences. Muse searches flights, hotels, and ground transport, presents options, and books once you confirm. No switching between Kayak, Google Flights, and hotel sites manually.

Form filling: it fills out forms across the web using your saved information. Permission slips, registration forms, applications. Tasks that feel trivial but eat time when they stack up.

Multi-step task planning: give Muse something bigger and it builds a plan, then advances it over time. "Set up a referral program for my business" becomes a series of steps Muse works through: researching tools, drafting the structure, setting up the mechanism while checking in with you at decision points.

Spawn sub-agents: for complex tasks, Muse creates multiple sub-agents that work in parallel on different parts of the goal simultaneously. One researches while another drafts while a third schedules. The parent agent coordinates.

How it works technically: every user gets their own dedicated cloud-hosted Linux virtual machine. Muse runs inside it with its own browser, its own ability to run scripts, and access to your connected apps. That's why it keeps working when you close the app. It's not running on your phone, it's running in the cloud on your behalf.

Connected services at launch: Gmail, Google Calendar, Ticketmaster, OpenTable, Stripe checkout, Spotify, and more. Although amazon restricted Muse's access citing terms of service, an early preview of the permission friction that will define the agent category going forward. Still many connected services are being added daily.

Muse is the most significant consumer AI product launch since ChatGPT. Not because it's the most capable AI in the world, but because it's the first agent product with Meta's distribution, Meta's price point (free), and Meta's simplicity: available to anyone with a phone and an email address. For small business owners, the immediate use cases are personal (managing your own inbox and calendar), but the implications for how customers interact with businesses are significant.

A freelance marketing consultant in Denver uses Muse as her personal operations layer. Before Muse, she estimated she spent 90 minutes a day on what she calls "admin orbits", the cluster of small tasks that surround actual client work.

Responding to scheduling emails. Booking travel for client visits. Filling out vendor registration forms for new client accounts. Researching and comparing SaaS tools before recommending them.

Two weeks after installing Muse on her phone, her workflow: she voice-notes a task to Muse in the Uber on the way to a meeting. "Book a hotel in Chicago for the week of October 14, under $200/night, near the Loop, cancel-friendly." Muse has it booked before the meeting starts.

She estimates 45–60 minutes saved daily on admin tasks that have nothing to do with her actual expertise. She hasn't changed a single thing about how she does client work. She just stopped doing the tasks that surrounded it.

"Should I trust an AI with access to my email and my payment info?"

This is a solid question and there's no single right answer, it depends on how you think about the trade-off.

Muse runs in a dedicated virtual machine with Meta's Sentinel oversight system, which independently monitors agent behaviour for anything outside the scope of what you asked.

Meta says it doesn't use the content of your emails to train its advertising models (interaction metadata trains the agent model, with an opt-out). On payments, Muse requires explicit approval before any transaction completes, it can't charge without a confirmation step from you.

The honest concern isn't catastrophic failure: it's the accumulation of small decisions made on your behalf that you didn't review.

The mitigation: start with low-stakes connections (calendar only), review the activity log daily, and only expand access to payment and email once you've seen how it behaves with less consequential tasks.

Delegate trust incrementally, the same way you would with a new human assistant.

The free tier is genuinely usable for light-to-moderate personal and business task delegation. The $20 Power plan is the right starting point for anyone who plans to use it daily. The $100 Maximum plan is for heavy users who want Muse running complex, ongoing multi-step tasks without hitting limits.

One important note: Meta has not published exact task counts or usage limits for each tier. "Higher usage limits" is the description, not a specific number. This is worth watching, like any agent product, heavy use of compute-intensive tasks could burn through limits faster than lighter use.

This week: install Muse, connect your Gmail, and delegate one task you were already going to do yourself in the next 48 hours. Not something hypothetical, something actually on your list.

A reply you owe someone. A booking you need to make. A form you've been putting off.

Do it through Muse instead of doing it yourself, and compare the outcome against what you'd have done.

That single test tells you more about whether Muse fits your workflow than any review will.

Muse hit 2.5 million downloads in its first two weeks, outpacing ChatGPT, Claude, and Grok at the same point after their respective launches.

It's currently the #1 free app on both the US iPhone and Android charts.

That adoption pace matters for small businesses for a reason that has nothing to do with using Muse directly: your customers are now downloading it too.

An AI that can browse, compare, and purchase on a user's behalf without them visiting your website or engaging with your marketing is a new kind of buying behaviour.

Businesses that are easy to discover and transact with through agent-mediated interfaces will have an advantage.

Businesses whose sales process requires a human-to-human interaction that an agent can't initiate may find a new friction point emerging.

Connecting Muse to every app and granting it maximum permissions before you've seen how it behaves on simple, low-stakes tasks.

The instinct with a new tool is to set it up completely and unlock everything at once.

With an AI agent that has access to your email, calendar, and payment methods, that instinct is worth resisting. Start narrow. Connect one app. Give it one category of task.

Watch what it does for two weeks. Then expand.

The agents that create problems for people in the first month are almost always the ones set up too broadly too fast, not because the AI is unreliable, but because the owner didn't yet know which tasks it could handle well and which it couldn't.

Until next issue, Kris

The Layman's AI — The only AI updates your business actually needs.