What Muse Is and Is Not

Before you set anything up, get the names straight. They sound alike, but each one has its own job and its own terms.

• Muse is Meta's personal AI agent, at muse.ai. An agent is software that carries out tasks for you, not just answers questions. Muse launched in the US on September 8, 2026. It was built by Meta Superintelligence Labs.
• The Meta AI app is a separate Meta product with its own terms. What you agreed to there does not govern Muse. Muse has its own Supplemental Terms at muse.ai/terms.
• Muse Spark is the model Muse runs on. A model is the AI engine underneath a product. Meta calls Muse Spark its most capable model for agentic work. You don't sign up for Muse Spark. You use Muse, and Muse uses Spark.

Unrelated products also share the name, such as Project MUSE journals and a "Muse" website builder. Check the address before you trust a page.

Why this matters at work: suppose your firm's AI policy approves "Meta AI". That approval doesn't automatically cover Muse. Ask IT about Muse by name.

You have probably used a chat assistant for one exchange at a time: paste in a report, get a summary, and that's the end of it. Muse works differently in four ways. Meta says it:

1. Carries out tasks, such as filling in forms or drafting and sending an email. In this course it asks for your approval first.
2. Keeps working on goals in the background after you close the app, either on a schedule or when something relevant happens.
3. Remembers what you tell it across chats. You can view that memory, edit it and ask it to forget.
4. Comes back to you when something changes or when it needs your approval.

Example from the sample project. The question is whether to bring a second supplier into the northern region before the contract renews. A chat assistant can summarise today's lead-time report. Muse can check every Monday whether that report or the quality team's figures have changed, and tell you only when they have.

That ability is also the risk. Work that runs while you're away needs tighter settings, and later lessons set them up.

Muse can work on its own because it has three tools a chat window lacks.

• Muse Secure VM. A VM (virtual machine) is a computer that exists in the cloud. Each user gets one, with a web browser, a file system and a terminal. Muse uses it to browse, store files and write small pieces of code for a task.
• Artifacts. These are outputs Muse makes and keeps: documents, PDFs, web pages, and interactive items such as trackers. Meta's help pages also list images, videos and podcasts.
• Connectors. These link Muse to other services. Meta names tools professionals use, including Google Calendar, Notion, Granola, Zoom, Asana, Slack and Box. The Mac app can also use local apps and files.

Example. For the supplier project, you could ask Muse for an Artifact that tracks open actions from your meeting notes. If your IT team has approved it, Muse reads those notes through a read-only Granola connector and builds the tracker on its VM. Then it waits for you.

Artifacts can be shared with a link. Lesson 5 covers why confidential ones should never be shared.

You can use Muse on:

• iOS and Android apps
• The web, at muse.ai
• Mac, through a desktop app launched in mid-September 2026. On Mac, Muse can use your computer directly. This course keeps that access tightly limited and never gives it the whole disk where client files are stored.
• WhatsApp, where you can message Muse directly

Meta has no official step-by-step help page for WhatsApp yet. Treat any unofficial WhatsApp number or link offering Muse as a scam. If a message promises "early Muse access" or bonus usage, don't tap the link. Open Muse only from muse.ai, the official app stores, or Meta's download page (ai.meta.com/muse/download/).

This matters more for professionals. Someone who fakes the sign-in could end up with an account you later connect to client calendars or project tools.

Screens and menus differ between platforms and change often. When this course names a menu, the version on your device may look slightly different.

At Meta Connect on September 23, 2026, Meta announced several Muse features. As of October 28, 2026, none of the following is confirmed as shipped:

• Voice mode
• Muse Realtime Avatar
• A Muse email address, which would let you cc Muse on email threads
• Muse on AI glasses, described as coming "in the coming months"
• Muse Confidential VM, which would encrypt your data with a key only you hold. Meta says "later this year".

This course doesn't teach or rely on any of them. Confidential VM matters most for client work. Under the current design, Meta restricts its staff's access by policy, but it can still access your data when needed to support, secure or operate the service. Plan around how Muse works today, not what has been announced.

Phone calls to businesses are also left out. Their public status is unconfirmed, and Meta rolled back a test of them.

Things change quickly, so before you rely on a feature, check Meta's newsroom and Help Center.

Here is the shift this course is built on. On Monday you get a one-line request: Should we bring a second supplier into the northern region before the contract renews? By Friday, a director has to be able to act on your answer without calling you to ask what it means.

Muse can do much of the middle: gathering evidence, summarising delivery history, tracking actions and drafting status notes. Two parts stay with you:

• The brief. Who acts on the answer, what decision it supports, what's excluded, and what "done" looks like.
• The handover. Every claim traced to a source, every commitment approved by a person, and nothing sent until you release it.

Meta's terms say the same: you are "solely responsible" for your instructions and for everything Muse does to carry them out.

Kent Beck, creator of Extreme Programming, has described working with AI this way: it "feels like an exoskeleton." He is still the programmer doing the programming, but he can take on work he couldn't have attempted alone. For you, that means bigger briefs, with you still the one vouching for the result.