Meta has handed its Muse AI agent three new abilities in the span of a single week: a real-time video avatar users can talk to face-to-face, a dedicated email address the agent can be looped into, and hands-on control of native Mac applications. The upgrades, unveiled around Meta Connect on September 23, 2026, mark the fastest capability expansion of any consumer AI agent since Muse itself launched just over two weeks earlier, according to reporting from The Decoder and The Verge.
The pace matters as much as the features. Meta introduced Muse on September 8, 2026, then shipped a Mac app roughly ten days later, and by September 23 was previewing avatars, agent-owned email, and smart-glasses access at its flagship hardware event. That cadence puts Meta in direct competition with OpenAI’s Operator, Anthropic’s computer-use agents, and Google’s Project Mariner, all of which have spent the past year chasing the same goal: an AI system that doesn’t just answer questions but actually operates your apps.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Meta Announced This Week
The headline change is that Muse is becoming a multimodal agent rather than a text-only chat assistant. According to The Decoder’s reporting on the announcement, Muse will soon support live video conversations through a customizable avatar, will receive its own email address so it can work in the background without the app staying open, and can already operate permitted Mac applications with user approval.
Of the three features, Mac control is the one users can test today. The video avatar and the dedicated email address were both described in the newest coverage as forthcoming, previewed at Meta Connect but without a locked public release date, per Zeniteq’s writeup of the announcement. That distinction is easy to miss in the headlines, but it shapes how the rollout should be read: one capability is shipping, two are being teased.
Meta also used the moment to expand Muse onto its AI smart glasses lineup, where the agent can reportedly be triggered hands-free with a wake word. Combined with the $1,299 VR glasses hardware Meta showcased at the same event, the company is positioning Muse as the connective layer across phone, desktop, and wearable form factors, rather than a single-app product.
Timeline: From Launch to Meta Connect in 15 Days
Muse’s rollout has moved unusually fast for a consumer AI product with this much access to personal accounts. Meta first introduced it as, in the company’s own framing, the first personal AI agent built to act on a user’s behalf, rolling out in the United States through iOS, Android, and the muse.ai website, according to Meta’s official announcement.
| Date (2026) | Milestone | Source |
|---|---|---|
| Sept. 8 | Muse launches in the US on iOS, Android, and muse.ai as a personal AI agent | Reuters, NYT, CNBC |
| Sept. 9 | Coverage confirms Muse can send emails, book trips, and make purchases through connected apps | Business Today |
| Sept. 17 | Meta ships a dedicated Muse mobile app for iPhone | 9to5Mac |
| Sept. 18-19 | Muse arrives on Mac, gaining opt-in access to files, messages, calendar, notes, and mail | TechCrunch |
| Sept. 23 | Meta Connect: live video avatars, a dedicated Muse email address, and smart-glasses access previewed | The Decoder, The Verge |
Reuters reported at launch that Muse was modeled on the open-source AI agent OpenClaw and designed to reach across a person’s connected apps in categories including email, calendar, payments, health, shopping, and smart-home controls. Users choose which apps the agent can touch and can revoke that access at any time, per Reuters’ original coverage of the launch.
Video Avatars: How Muse’s Real-Time Face Works
The video feature is powered by what Meta is calling a Muse Realtime Avatar model, which governs how the avatar looks and moves during a live conversation, according to reporting cited by The Decoder and The Verge. Rather than a static profile picture, the avatar is meant to behave like a visible presence: users will be able to see Muse “at work,” similar to how a colleague might visibly sit at their desk.
CNN’s early hands-on account of the feature, published September 23, described a “Pixar-like avatar” that appeared to type away on a laptop at the top of the screen while the reporter’s AI helper worked in the background, based on CNN’s reporting on testing the tool.
Customization extends past appearance. Users will reportedly be able to name their agent, design its avatar, and adjust voice characteristics such as speaking pace or accent, according to coverage from LiveMint. That level of personalization echoes what Meta already allowed at Muse’s original launch, when users could name the agent and pick a look during onboarding, per Business Today’s coverage of the September 9 rollout.
Why a Face Changes the Product
Putting a rendered face on an agent is not a cosmetic choice. It changes how the tool gets used. A text-based assistant lives inside a chat window that users open and close. A video avatar, by contrast, is designed to feel present and ambient, more like a coworker sharing a screen than a search box. That shift toward “ambient” AI is part of a broader industry trend, visible in how Meta positions Muse alongside its smart glasses and VR hardware rather than as a standalone app.
Muse Gets Its Own Email Address
The second new capability lets users email Muse directly, add it to an existing email thread, or forward messages for the agent to act on, according to reporting from The Decoder and The Verge. The intent, per that coverage, is to let Muse operate in the background on ongoing tasks rather than requiring someone to keep the Muse app open every time they want the agent to do something.
This mirrors a pattern other AI companies have tested this year. Giving an agent its own inbox turns email into a control channel, not just a task the agent completes. It is conceptually close to how ChatGPT’s voice assistant recently added email, calendar, and Slack access, treating messaging platforms as the interface layer for agent instructions rather than a destination the agent merely reads and writes to.
Meta has not published a fixed date for when the email feature becomes broadly available, and the current reporting frames it as a preview rather than a shipped product, consistent with Zeniteq’s account of the Meta Connect announcements.
Mac Control: What Muse Can Actually Do on a Computer
Unlike the avatar and the email address, Mac control is live now. TechCrunch reported that Muse’s Mac app lets the agent interact with a user’s files, messages, calendar, notes, and mail, working inside their native applications rather than confined to a browser tab, according to TechCrunch’s coverage of the rollout.
Access is opt-in at the application level, and Meta says Muse asks for user approval before completing sensitive actions, per the same TechCrunch report. That approval-gated design is the same basic safety pattern used by other computer-use systems: the agent can see and act, but a human is meant to stay in the loop for anything consequential, such as sending a payment or deleting a file.
9to5Mac’s coverage of the September 17 iPhone app launch noted that Muse is also reachable through the web at muse.ai on Mac, and that the customization options for agent name and avatar carry over across platforms, according to 9to5Mac’s reporting.
Smart Glasses and the Shopping Push
Meta Connect also extended Muse to the company’s AI smart glasses lineup, where the agent can reportedly be invoked hands-free with a wake word, per Zeniteq’s summary of the event. That puts Muse in the same product family as Meta’s recently unveiled 100-gram VR glasses, which the company priced at $1,299.99 and positioned as a rival to Apple’s Vision Pro line.
On the commerce side, LiveMint reported that Muse’s expansion includes shopping support through partners such as Shopify, PayPal, Walmart, and Sephora, suggesting Meta intends the agent to complete real purchases rather than just recommend products. Combined with payments access already part of Muse’s original app permissions, per Reuters’ launch-day report, that positions Muse as a transactional agent, not just a conversational one.
How Muse Stacks Up Against Rival Computer-Use Agents
Meta is not the first company chasing an agent that operates a computer on a user’s behalf. OpenAI’s Operator focuses on carrying out tasks inside a web browser. Anthropic’s computer-use capability lets Claude models interact directly with computer interfaces, including clicking, typing, and navigating applications. Google’s Project Mariner targets browser-based task execution in a similar vein. What the newest Muse announcements add to that category, per the combined reporting from The Decoder, The Verge, and TechCrunch, is persistent communication through a dedicated email address, real-time avatar interaction, smart-glasses access, and native Mac-app control, rather than browser automation alone.
| Agent | Maker | Primary Surface | Standout Capability (per current reporting) |
|---|---|---|---|
| Muse | Meta | iOS, Android, Mac, smart glasses | Video avatar, dedicated email address, native Mac-app control |
| Operator | OpenAI | Web browser | Autonomous browser task completion |
| Computer Use | Anthropic | Desktop interface (via Claude) | Direct interface interaction: clicking, typing, navigating apps |
| Project Mariner | Web browser (Chrome) | Browser-based multi-step task execution |
Notably, none of the current reporting on this week’s Muse announcement includes a standardized, third-party benchmark showing Muse is faster, more accurate, or more autonomous than these competing systems. The comparison at this stage is about breadth of surface area (phone, desktop, glasses, email, and avatar in one product) rather than a head-to-head performance claim.
Why Meta Is Moving This Fast
The speed of Muse’s rollout looks less like an accident and more like a deliberate land grab, and Meta had already laid model-level groundwork for it: on September 2, 2026, Meta AI Research released Muse Spark 1.3, an update aimed specifically at improving reasoning and agentic-task performance, the two areas most relevant to an agent that is supposed to act on a user’s behalf. Every major AI lab is racing to define what a “personal agent” means to consumers before the category settles into fixed expectations, and Meta has an obvious structural advantage: it already owns the identity graph, the messaging rails, and now the hardware (phones aside) through Ray-Ban and Oakley smart glasses plus its own VR headsets. An agent that can act across all of that is a different product than a chatbot bolted onto a search bar.
There is also a defensive angle. CNBC’s coverage of the original launch framed Muse’s arrival within what it called a “public reckoning” over AI agent safety and privacy, noting that Meta pushed into personal AI agents even as scrutiny of the category intensified, according to CNBC’s reporting. Shipping fast while competitors are still iterating gives Meta first-mover data and habit formation, even if it also means shipping into a more skeptical regulatory and public mood.
The Privacy and Security Questions Still Unanswered
An agent with a Mac control, an email address, and payment access is also a bigger attack surface than a standard chatbot, and Meta’s own product has already run into trouble on this front. Reporting on Muse’s security posture found that a backdoor tied to the agent went unpatched for at least 24 hours after disclosure, and separate coverage detailed a zero-day that exposed more than 50 hidden commands through a single misconfigured setting. Amazon reportedly moved to block Muse from its own AI agent ecosystem within roughly two weeks of the agent’s public debut.
None of that history is mentioned in this week’s feature announcements, but it is the backdrop against which they land. Adding native Mac access, a standing email address, and an always-on video presence multiplies the number of places something could go wrong, even with opt-in permissions and approval prompts in place. Meta’s public safeguards, per TechCrunch’s reporting on the Mac rollout, rest on the idea that sensitive actions require explicit user approval and that app-level access can be revoked at any time, echoing the same permission model Reuters described at launch.
Historical Context: From Chatbots to Agents With Hands
Muse’s evolution traces the same arc the entire AI industry has followed over the past two years: text chatbot, then multimodal assistant, then an agent that can take action, and now an agent with a persistent identity across email, video, and hardware. Each step has compressed. Where earlier chatbot upgrades were measured in quarters, Muse went from a September 8 launch to a Mac app, an iPhone app, live avatars, and its own email address inside about fifteen days, while Meta for Developers separately pushed its companion coding model, Muse Code, out of beta that same September for long-horizon engineering tasks, a sign that Meta’s broader AI stack was maturing in parallel with the consumer-facing app.
That pace is only possible because the underlying building blocks, screen-reading models, voice synthesis, and app-permission frameworks, were already mature from earlier computer-use agent efforts across the industry. Meta is not inventing the category so much as packaging it faster and distributing it wider than rivals, using its existing app install base on iOS and Android to skip the slow bootstrap phase that new AI products usually face.
Market Impact and What Analysts Are Watching
For Meta, Muse is a proving ground for whether its AI investment translates into daily consumer habit rather than novelty use, and the company has already started testing how to monetize that habit: as of October 2026, Muse remains free with usage limits for casual users, while Meta has rolled out $20-per-month and $100-per-month subscription tiers for people who want higher usage caps, according to CNBC. The company has already tied Muse to its hardware roadmap, pairing the agent’s expansion with the debut of its 100-gram VR glasses at $1,299 and continued Ray-Ban smart glasses integration. If Muse becomes the default way people manage email, shopping, and calendars across devices, it strengthens Meta’s position against Apple, Google, and OpenAI in the fight over who owns the “default assistant” layer of a person’s digital life.
For competitors, the pressure is now on to match feature breadth, not just model quality. OpenAI, Anthropic, and Google have each shipped narrower computer-use tools focused on browser and desktop automation; none has yet combined that with a persistent email identity and a live video avatar in one consumer product. Whichever company closes that gap first will shape how the next wave of AI agent regulation gets written, since lawmakers and privacy regulators are watching the permission models these tools use as a template for future rules.
Rollout Status at a Glance
| Feature | Status as of Sept. 25, 2026 | Platform |
|---|---|---|
| Core Muse agent (email, calendar, payments, shopping, smart home) | Live | iOS, Android, muse.ai |
| Mac app with native app control | Live | Mac |
| Dedicated Muse email address | Previewed, not yet broadly available | Cross-platform |
| Real-time video avatar (Muse Realtime Avatar model) | Previewed, no confirmed public date | Cross-platform |
| Smart-glasses wake-word access | Announced at Meta Connect | Meta AI smart glasses |
Predictions: Where Muse Goes From Here
Based on the pace and pattern of Meta’s rollout so far, a few directions look likely over the next two to three months. First, expect the dedicated email address to ship before the video avatar, since email infrastructure is simpler to stand up than a real-time rendering pipeline, and Meta has already proven it can move fast on account-level features. Second, expect scrutiny of Muse’s Mac permissions to intensify, particularly given the agent’s existing security history; regulators and security researchers are likely to test the approval-prompt safeguards TechCrunch described.
Third, expect competitors to respond with their own multimodal pushes. Given how quickly OpenAI, Anthropic, and Google have shipped computer-use features this year, a rival avatar-based or persistent-identity agent announcement within the next quarter would not be surprising. Fourth, expect Meta to lean harder into the smart-glasses tie-in, since hands-free wake-word access to an agent that can already touch email, calendar, and shopping is a natural fit for a wearable form factor Meta has already committed hardware dollars to. Fifth, expect the shopping integrations with partners like Shopify, PayPal, Walmart, and Sephora to expand, since transactional use cases are the clearest path to justifying Muse’s broad permission requests to a skeptical public.
What This Means for Users Right Now
For anyone already using Muse, the practical takeaway is straightforward: the Mac control feature is real and usable today, with opt-in, app-by-app permissions and approval prompts for sensitive actions. The video avatar and the dedicated email address are not yet broadly available, so claims about “talking to Muse on video” or “emailing your agent” should be treated as previews rather than shipped features until Meta confirms a public release window.
Anyone weighing whether to grant Muse deeper access, particularly Mac file and mail permissions, should factor in the agent’s documented security history alongside its expanding capabilities. Broader permissions and a broader attack surface tend to move together, and Muse’s last two weeks illustrate both sides of that trade-off: fast, genuinely useful feature growth, paired with the kind of access that makes any single misconfiguration more consequential.
Frequently Asked Questions
What is Meta Muse?
Muse is Meta’s personal AI agent, launched September 8, 2026, in the US on iOS, Android, and the muse.ai website. It is designed to act on a user’s behalf across connected apps, including email, calendar, payments, health, shopping, and smart-home controls, according to Reuters’ coverage of the launch. As of October 2026, Muse is free to use with usage limits, and Meta also offers $20-per-month and $100-per-month subscription tiers for users who want higher caps, per CNBC.
Can Muse control my Mac right now?
Yes. TechCrunch reported that Muse’s Mac app, live since mid-September 2026, can interact with files, messages, calendar, notes, and mail inside their native applications, on an opt-in, app-by-app basis with approval prompts for sensitive actions.
Is the Muse video avatar available yet?
Not broadly. It was previewed at Meta Connect on September 23, 2026, powered by a new Muse Realtime Avatar model, but current reporting from The Decoder and The Verge does not indicate a confirmed public release date.
How does the Muse email address work?
Once available, users will reportedly be able to email Muse directly, add it to an email thread, or forward messages for it to act on, letting the agent work in the background without the app staying open, per The Decoder’s reporting.
How is Muse different from OpenAI’s Operator or Google’s Project Mariner?
Operator and Project Mariner are primarily browser-task agents. Muse combines native Mac-app control, a persistent email identity, a real-time video avatar, and smart-glasses access in one product, though no standardized benchmark currently compares their performance directly.
Is Muse safe to give access to email and files?
Meta says access is opt-in and gated by approval prompts for sensitive actions. However, the agent has a documented security history, including a reported unpatched backdoor and a separate zero-day exposing dozens of hidden commands, which is worth weighing before granting broad permissions.
What shopping partners work with Muse?
LiveMint reported that Muse’s expanded shopping capabilities include partners such as Shopify, PayPal, Walmart, and Sephora.
Will Muse work with Meta’s smart glasses?
Yes. Meta announced that Muse is expanding to its AI smart glasses lineup, where it can reportedly be triggered hands-free using a wake word, alongside the broader hardware push that includes Meta’s newly unveiled VR glasses.
Related
- Meta Muse Backdoor Stays Unpatched 24 Hours Later
- Meta Muse Zero-Day: 1 Setting Exposes 50+ Commands
- Amazon Blocks Meta Muse AI Agent After 12 Days
- Computer-Use AI Agent Setup: 13 Steps, 90 Min
- ChatGPT Voice Adds Email, Calendar, Slack Access
- Meta VR Glasses Debut at 100g, $1,299.99 Price
Related Coverage
- Amazon Blocks Meta Muse AI Agent After 12 Days [2026]
- How qubit control and readout make quantum computing work
- ChatGPT Voice Adds Email, Calendar, Slack Access [2026]
- OpenAI Claims 100+ Math Problems Solved in 24 Days [2026]
- Muse Spark Tops GPT-6 Astra, Claude Fable: 75% SWE [2026]
