What Muse Is, and What It Isn't

Meta uses the word "Muse" in more than one place, and the difference affects your settings later.

• Muse is Meta's personal AI agent. It launched on September 8, 2026, and you reach it at muse.ai, in the iOS and Android apps, or in the Mac app. An agent is an AI that carries out tasks rather than only answering questions.
• Meta AI is a separate Meta app with its own terms. Settings you change in Meta AI don't change Muse, and the reverse is also true.
• Muse Spark is the model that Muse runs on: the engine, not the product you open.

Other unrelated products also use the name (a website builder, the Project MUSE journal library, and more). When you search for help, make sure the page is about Meta's agent at muse.ai.

Example: you read a post saying "Muse Spark can now book trips." That is really about the Muse product. The model doesn't book anything by itself. When in doubt, ask: is this the app I open, the model behind it, or Meta's other assistant?

Mark Zuckerberg, Meta's CEO, spoke about this difference on the Sources Podcast in September 2026, recorded at Muse's launch. His point, in short: with Meta AI, ChatGPT or Gemini you send one prompt and get one answer. With an agent, "you give it goals", and it keeps working until they're met.

Here is what that looks like on a channel.

Single prompt: "Write a caption for my cliff-walk Reel." You get a caption, and the job is done.

Standing goal: "Every Monday at 9am Eastern, take the comments I've exported (usernames removed), group the questions by what viewers need, merge duplicates, attach my earlier answer where there is one, and give me a draft list. Don't reply to anyone."

The second one keeps running. Muse returns to it each week without being asked again, and it checks with you at the points you named.

A standing goal is a recurring job you hand over once, with a scope, sources, a format and stop points. The agent keeps working on it on a schedule or when something changes. A single prompt is one request and one answer.

Muse can keep working after you close the app because of five parts. You'll use all of them in this course.

• Muse Secure VM: its own cloud computer, with a web browser, files and a terminal. It keeps running when your phone is off.
• Connectors: links to outside services. Meta names Gmail, Google Calendar, Instagram professional account analytics, Canva and Notion, among others. You'll connect them one at a time later.
• Memory: what Muse keeps about you between chats, stored in a file you can open and edit (Assistant icon > Identity > Memory).
• Artifacts: the things it makes, such as documents, PDFs, web pages, trackers and checklists. Your weekly question list would be an Artifact.
• Approvals: before important actions, Muse shows you an approval card. How often it asks depends on Settings > Permissions. Engadget reports that the default, Ask for some actions, asks only for some actions. Always ask asks before every action, and that's what you'll set.

Menu names come from Meta's help pages. The app changes often, so your screen may look a little different.

Here is the Monday comment job from step 2, followed from start to finish.

On Sunday you set the goal in chat: scope (questions from last week's export), format (a table grouped by viewer need), and a stop point ("don't contact anyone; ask me before using any new website").

On Monday morning Muse works on its cloud computer while you film. It merges three versions of "when's the last shuttle back?" into one item and attaches the answer you gave in the video.

The result is an Artifact, a draft list. If the job needs something your settings say it must ask about, an approval card appears first. You tap See task details before you decide.

You edit the list and choose what to answer. Muse never posts the answers.

Once a week you open the Activity log, a time-ordered record of what Muse did and what it was allowed to do, and check that it matches what you asked for.

Put the steps in order.

Muse is built to act for you. Meta's launch materials mention sending email, filling in forms, buying and negotiating. This course deliberately uses only part of that.

Course rule: Muse drafts and you approve. Nothing is sent, posted or bought without your go-ahead. Muse never makes or changes your face or voice, including spoken lines in your videos. That is stricter than some platform rules, and it's on purpose: your audience books trips and buys gear on your word.

There are two reasons for this. Meta itself says Muse "will sometimes make mistakes", and Muse's terms make you responsible for everything it does on your instructions. A widely reported case shows the risk: a YouTuber gave Muse a blanket "always" permission on Facebook Marketplace chats, and it gave a buyer his home address.

Announced but not confirmed as shipped (as of October 28, 2026): voice mode, Realtime Avatar, a Muse email address, phone calling and Muse on AI glasses. The course doesn't teach any of them. If one appears in your app later, check Meta's official pages before relying on it.

This course's line is stay the author. Here is how that works on one question.

Under Sunday's video, a parent asks whether their children, six and eight, can manage the cliff walk next month, and how to get back to the car. Forty other people are wondering the same thing. You walked it in May: it's gentle until the last mile of 300 steps, your nephew sat down on the 200th, and the last shuttle leaves the far end at 5:30.

Muse can do the legwork: sort the forty comments, pull the park's timetable page with a link and date, and draft the TikTok, Reel and YouTube versions. You keep the walk, the word and the final check. That means the 200th step, the decision that 5:30 is right, and the go-ahead to post.

Jim Louderback, former CEO of VidCon, suggests a test: look at your last twenty posts and ask which ones an AI could have made. Use a quicker version here. List the prep jobs you repeat every week, the ones that don't need your experience or your voice, and you have your first agent jobs.