Level 1 — Absolute Beginner
Meta is a very big technology company. It owns Facebook, Instagram and WhatsApp. In September it made a new app called Muse.
Muse is a helper on your phone. It can write emails for you. It can book a hotel or a flight. It can fill in forms.
Muse remembers what you like. It keeps working when you close the app. You can use it on a phone or on a computer.
Many people downloaded it very fast. This week it became the number one app in the United States. Meta shares went up.
- technology
- machines and computers made with science
- app
- a small program you use on a phone or tablet
- helper
- someone or something that helps you
- email
- a message you send on a computer or phone
- book
- to arrange a room, a seat or a ticket in advance
- form
- a paper or screen with questions you must answer
- remember
- to keep something in your mind and not forget it
- download
- to copy something from the internet onto your device
Level 2 — Elementary
Meta launched Muse on September 8, 2026, and described it as a personal AI agent rather than a chatbot. The difference matters. A chatbot answers your questions, while an agent is supposed to go away and actually do the task for you.
Muse can send email, book travel, fill in online forms and even negotiate on a user's behalf. It can open a web browser by itself, it remembers preferences you mention only once, and it carries on working after you close the app. Payments go through Stripe's Link service, and Meta says Shop Pay will be added soon.
Because an agent with that much freedom is risky, Meta built several protections. Muse runs on a separate virtual machine called Muse Secure VM that other programs cannot reach. A second program, the Sentinel agent, has to approve anything Muse sends out to the internet, passwords are stored in encrypted form so that Muse never sees them, and users set permissions app by app.
The reception has been fast. On Tuesday, September 22, Muse reached the top position on Apple's App Store in the United States, and Meta shares rose again after climbing almost 12 percent the previous day. The app is free for basic features, with paid subscriptions for heavier use, and Meta says a version for its AI glasses is on the way.
- launch
- to make a new product available to the public
- agent
- a program that carries out tasks for a person
- negotiate
- to discuss something in order to reach an agreement
- browser
- a program used to look at pages on the internet
- preference
- something you like better than the alternatives
- virtual machine
- a computer that exists as software inside another computer
- encrypted
- turned into a secret code so that others cannot read it
- subscription
- a regular payment to keep using a service
Level 3 — Intermediate
Meta released Muse on September 8, and by Tuesday, September 22, it sat at number one on Apple's App Store in the United States, a fortnight that has done more for the company's share price than most of its hardware launches. Investors pushed the stock up almost 12 percent on Monday and higher again on Tuesday, treating the app as proof that Meta has found a consumer product for the agent era.
The pitch separates Muse from the assistants that preceded it. Where a chatbot returns text, Muse is designed to act: it drafts and sends email, books travel, completes forms, negotiates on a user's behalf, converts stated goals into step by step plans, and volunteers suggestions based on preferences it has inferred. It opens a browser when it needs one, remembers a detail mentioned once, and continues executing after the user has closed the app. Purchases run through Stripe's Link, with Shop Pay promised later.
Autonomy of that kind creates an obvious attack surface, and Meta has answered it with architecture rather than assurances. Muse runs inside a dedicated virtual machine, Muse Secure VM, isolated from other agents. A separate Sentinel agent reviews and approves every action bound for the open internet. Credentials are held encrypted and never exposed to the model itself, sensitive steps require explicit user approval, and permissions are granted application by application. A fully end to end encrypted Muse Confidential VM is promised.
Underneath sits Muse Spark, which Meta calls its most capable model to date and describes as purpose built for real world agentic work. The company is distributing the app on iOS, Android and the web in the United States first, with support for its AI glasses to follow, and is charging nothing for basic use while reserving heavier workloads for paid tiers. Whether consumers will hand a commercial agent their inbox, their calendar and their card details remains the open question.
- fortnight
- a period of two weeks
- pitch
- the argument a company makes for why you should want its product
- infer
- to work something out from evidence rather than being told it directly
- autonomy
- the ability to act independently without being controlled
- attack surface
- all the points where a system could be broken into
- isolated
- kept separate from everything around it
- credentials
- the usernames, passwords and keys that prove who you are
- tier
- one level in a set of ranked prices or services
Level 4 — Advanced
Meta shipped Muse on September 8 and watched it take the top slot on Apple's American App Store fourteen days later, a trajectory that has moved the company's equity more decisively than any headset it has ever sold. The stock rose almost 12 percent on Monday and extended those gains on Tuesday, which is less a verdict on the software's quality than on the market's conviction that Meta has located a mass market use case for agentic AI before its rivals did.
The product's premise is a departure from the conversational paradigm. A chatbot terminates in language; Muse is built to terminate in outcomes. It composes and dispatches correspondence, arranges travel, completes forms, negotiates on the user's behalf, decomposes stated ambitions into sequenced plans, and offers unprompted suggestions derived from preferences it has inferred rather than been given. It instantiates a browser when a task demands one, retains a detail disclosed a single time, and persists after the application has been dismissed, with settlement routed through Stripe's Link and Shop Pay signalled for a later release.
Delegation on that scale converts a consumer app into a security problem, and Meta's response is notable for being structural rather than rhetorical. Muse executes within a dedicated Muse Secure VM that other agents cannot reach; a discrete Sentinel agent adjudicates every outbound action before it touches the public internet; credentials sit encrypted and are never surfaced to the model; sensitive operations demand affirmative user consent; and authority is scoped application by application. An end to end encrypted Muse Confidential VM is on the roadmap.
Beneath the interface runs Muse Spark, which Meta characterises as its most capable model yet and as purpose built for real world agentic work, distributed across iOS, Android and the web in the United States with AI glasses support to follow. The commercial architecture is familiar enough, a free tier widening the funnel and paid plans capturing intensity of use. The unresolved variable is trust. An agent useful enough to be worth having is, by construction, an agent holding the keys to a user's inbox, calendar and payment instruments, and no amount of isolation quite dissolves that bargain.
- trajectory
- the path something follows as it develops over time
- equity
- the shares that represent ownership of a company
- paradigm
- a standard model or pattern of doing something
- decompose
- to break something down into smaller parts
- instantiate
- to create an actual working copy of something
- adjudicate
- to judge or decide formally on a matter
- scoped
- limited to a defined range or set of permissions
- funnel
- the path by which potential customers become paying ones
Talk about this article with KatePractice this story's vocabulary in a real spoken conversation with your AI tutor