What Muse is and what it is not

This course teaches you to use Meta's Muse on the standing work of a finance team: variance reviews, the close calendar, payout reconciliations and first-draft commentary. One rule applies throughout:

AI can produce the number. Only you can stand behind it.

Muse can return a variance table in minutes. It can't sign the pack. Before any figure it prepares leaves finance, you tick it back to source, the same way you would check a schedule a new hire built.

Richard Jackson, EY's global AI assurance leader at the time of his talk, suggested a useful way to think about this. Treat the tool as a "first-year equivalent". The profession already knows how to review a first-year's work: a senior does a detailed review, then a manager does a general one. In this course, every Muse output gets that kind of review.

Example: Muse reports that travel expense rose 38% month over month. Before that goes into commentary, you pull both months from the ledger, recompute the $ change and the %, and confirm the period and currency. If your figure doesn't match, nothing ships.

The track's tagline sums it up: Sign only what ties out.

Several products share the name, so get the right one before anything else.

• Muse (muse.ai) is Meta's personal AI agent. Meta Superintelligence Labs built it and it launched on September 8, 2026. When you open muse.ai, it sends you to a Meta sign-in. That's the product this course covers.
• Muse Spark is the model Muse runs on. A model is the underlying AI engine. Muse Spark isn't something you set up or log into here.
• Meta AI is a different Meta app with its own terms. Settings, permissions and data rules you change in Meta AI don't carry over to Muse, and Muse's don't carry over to Meta AI.
• Unrelated products use the name too, such as Project MUSE (a database of academic journals) and a website builder called Muse. They have nothing to do with Meta.

Why it matters in finance: your IT team may already have a view on "Meta AI". That view may not cover Muse, which has different terms, connectors and data handling. When you ask for approval, name the product exactly: Meta's Muse agent at muse.ai.

You may have tried a chat assistant: you ask a question, it answers, and nothing happens until you ask again. An agent works differently. It carries out tasks. Meta's examples include sending an email, filling in forms, making purchases and booking travel. It also keeps working toward a goal after you close the app, and comes back when something changes or when it needs your approval.

For a finance team, that changes the risk.

Example: you ask a chat assistant why office supplies rose 40%. At worst you get a wrong paragraph. You ask an agent connected to your books to "clean up the office supplies category". If you gave it the access, it could try to recategorize transactions in your system of record.

That's why this course limits Muse to reading and preparing. Muse will not post entries, pay bills, move money or send anything without your approval. The access you grant and the approvals you give are the controls, much as segregation of duties is a control in your close. Later lessons set those controls up before anything gets connected.

Each user gets a Muse Secure VM, a private virtual computer in Meta's cloud. It has a web browser, a file system and a terminal. Muse uses it to browse, store files, write and run code, and build tools for a task.

What it produces are called Artifacts: documents, PDFs, web pages and interactive outputs such as trackers. In this course you'll ask for a close-tracker Artifact and a draft scenario Artifact.

A long-running task usually goes like this:

1. You give Muse a goal.
2. It plans the steps.
3. It works on them in the background, even after you close the app.
4. If it needs your approval, it stops and shows you an approval card.
5. It returns with a result, often an Artifact.
6. You check the result against source.

Example: you ask for a tracker of your BD1–BD5 close steps. Muse builds it on its VM and hands it back. You still check that every step and owner from your actual checklist made it in.

One caution for later: Artifacts can be shared by link, and anyone with the link can open them. In October 2026 Meta paused sharing after shared-page addresses turned up in public DNS records. This course never shares Artifacts that contain financial data.

Connectors link Muse to other services so it can read data and, if you allow it, act there. Meta's own lists include several that finance teams use:

• Intuit QuickBooks: the ledger behind a variance review or reconciliation
• Stripe and Shopify: payment processor payouts
• Gmail: vendor invoices arriving by email

The Mac app goes further. With your macOS permissions, it can use local apps (Mail, Notes, Calendar, Messages, Reminders) and files on your computer.

Example: to reconcile Stripe payouts to deposits, Muse would need to read Stripe and QuickBooks. It doesn't need to write to either. Wherever a connector offers a retrieval-only (read-only) setting, this course uses it.

Two things to know now and practise later. First, what each connector can read or do varies, so check its permission screen before you rely on it. Second, connecting a work account needs your employer's or IT team's approval first. You'll connect one service at a time, in a later lesson, after setting permissions.

These are the places you'll use most. Names come from Meta's Help Center, but the app changes often, so your screen may differ. If a label doesn't match, look for the closest equivalent or ask Muse where to find it.

• Chats tab: one long main chat for ongoing work, plus side chats that keep a project's context separate. You might keep the monthly close in its own side chat.
• Goals tab: long-running aims and the plans Muse builds for them.
• Assistant icon > Upcoming: scheduled tasks and reminders. This is where you'll confirm your BD1–BD5 close reminders say what you meant.
• Assistant icon > Activity log: a record, in time order, of what Muse did and which permissions it used.
• Assistant icon > Identity > Memory: what Muse has remembered about you.
• Settings: including Permissions, Connectors and Data controls.

Example: on BD1 you ask in a side chat, "Remind me at 9 am on each close day which steps are due." Afterwards you open Upcoming to check that five reminders exist, at the right times.

At its Connect event on September 23, 2026, Meta announced several Muse features: a voice mode, a Muse Realtime Avatar, a Muse email address you could cc on threads, and Muse on AI glasses. Meta gave no ship date for the first three and said the glasses would come "in the coming months". Press has also reported a phone-calling feature, and its public status is unconfirmed.

This course teaches none of them. If you read a guide that relies on any of these, treat it as unverified until Meta's own pages say the feature is live where you are.

The same goes for one more item: a WhatsApp connector. Officially, WhatsApp is somewhere you can message Muse. A connector that reads your WhatsApp hasn't been confirmed.

Example: a colleague suggests you "just cc Muse" on the vendor-statement thread. As of October 28, 2026, Meta hasn't confirmed that the Muse email address has shipped. In this course, invoices from email are collected through the read-only Gmail connector instead.

Features change quickly, so check Meta's newsroom and Help Center before you plan around any of them.