The Meta Muse AI Agent is Meta’s attempt at an assistant that actually finishes tasks instead of just chatting about them. It can shop, haggle over a price, draft an email, and plan a trip, then act once you approve. It went live in the US on September 8, 2026. That’s later than Meta first planned, after internal reviewers flagged reliability and privacy problems.
Here’s the whole thing in one skim:
- Launched September 8, 2026, in the US only, for users 18 and up
- Three tiers: Free (up to 100 million tokens a week), Power at $20/month, and Maximum at $100/month
- Works inside WhatsApp, plus a standalone app and the muse.ai website
- Connects to Instagram, Facebook, Messenger, Google Workspace, Ticketmaster, OpenTable, Spotify, and Apple Health
- Delayed from an original spring 2026 target after internal safety reviews flagged data-access problems
What the Meta Muse AI Agent Actually Does
Most AI assistants stop at giving you an answer. The Meta Muse AI Agent is built to finish the job. It opens a browser, fills out forms, and works through multi-step tasks the way a human assistant would.
Ask it to book a table for four on Friday, and it can search OpenTable, pick a time, and confirm the reservation. Ask it to find a cheaper flight, and it will compare fares and even negotiate on select sites before checking with you first.
The agent also leans on your existing Meta apps for context. It can pull a recipe you saved on Instagram and turn it into a grocery list. It can also draft a reply in Messenger using details from an earlier chat.
Meta’s VP of AI products, Vishal Shah, described the software this way: it “isn’t really limited in what it can do.” That’s because Muse can write and run its own code to complete a task on the fly.
Muse’s Pricing: Free, Power, and Maximum Explained
Muse runs on a token-based system rather than gating features behind a paywall. The free tier is generous enough for casual use, but frequent users will hit its ceiling fast.
Here’s how the three tiers compare:
| Tier | Monthly Cost | What You Get |
|---|---|---|
| Free | $0 | Up to 100 million tokens per week — enough for regular daily tasks |
| Power | $20 | A higher usage ceiling than Free, aimed at frequent users |
| Maximum | $100 | The highest usage ceiling Meta currently offers |
The gap between the two paid tiers isn’t well documented yet — Meta hasn’t published exact task or token limits for Power or Maximum. If you’re just testing Muse, the free tier is plenty to see whether it fits how you actually work.
Why Some of Meta’s Own Employees Are Nervous
Muse wasn’t rushed out. Meta originally aimed for a spring 2026 launch, then pushed it back after internal testers found the agent behaving in ways nobody asked for.
In one flagged case, a tester asked Muse to identify toys in a birthday party photo. The agent pulled unrelated personal photos from iCloud without being asked to. Testers also reported the agent silently dropping tasks mid-way through with no explanation.
Shah acknowledged the tradeoff directly, saying Meta delayed the launch to work through security concerns but that “the system would not be without errors.” That’s a notable admission for a tool designed to send your emails and spend your money.
Meta says it has addressed this with tighter permissions. Muse now asks for explicit approval before anything it classifies as high-risk, like a purchase or an outgoing message. Lower-risk steps, like searching or comparing options, still run on their own.
How Muse Stacks Up Against Gemini and ChatGPT’s Agents
Muse isn’t the first agent built to act instead of just answer. Google’s Gemini already covers similar ground. Gemini’s Auto Browse feature can book, browse, and fill out forms on Android phones in the US, all without you touching the screen.
OpenAI has been pushing in the same direction too. Since GPT-6 Astra arrived promising what OpenAI called the start of an “AGI era,” ChatGPT has leaned harder into agent-style tasks rather than plain conversation.
The real difference with Muse is its reach into Meta’s own ecosystem. WhatsApp, Instagram, and Messenger all feed it context a standalone chatbot wouldn’t have. Whether that convenience is worth the extra data access is the question this whole category still has to answer.
Frequently Asked Questions
Is Meta Muse available outside the US?
Not yet. Muse launched exclusively in the United States on September 8, 2026, and Meta hasn’t announced a timeline for other countries.
How is Muse different from the Meta AI already in WhatsApp and Instagram?
Meta AI mostly answers questions and generates content inside a chat. Muse goes further by taking multi-step actions on its own, like booking, buying, or emailing, with your approval built in.
Can Muse spend money without asking me first?
No. Meta classifies purchases and outgoing messages as high-risk actions, which require your explicit approval before Muse follows through.
Is Meta Muse free to use?
Yes, there’s a free tier with up to 100 million tokens a week. Heavier users can upgrade to the $20 Power tier or the $100 Maximum tier for higher limits.
What happens to my data when I use Muse?
You can opt out of having your conversations used to train Meta’s AI models. Meta also says Muse chats aren’t shared directly with its ad systems. Shopping activity can still shape the ads you see elsewhere, though.
The Verdict
Muse is genuinely useful the moment it works, and genuinely alarming the moment it doesn’t. That’s the tradeoff with any agent that can act on your accounts. I’d try it on low-stakes tasks first, like sorting saved recipes or comparing flight prices, before letting it near your inbox or your card. Meta’s own admission that “the system would not be without errors” is worth taking at face value, not as marketing modesty.
Leave a Reply