Meta’s New AI Agent Doesn’t Just Answer You, It Goes Off And Does The Work

Chatbots have gotten very good at talking. What most of them still can’t do is actually go do something on your behalf — book the flight, fill out the form, negotiate the bill down. Meta thinks it’s cracked that next step, and on September 8, 2026, it launched Muse, a personal AI agent built specifically to act rather than just respond.

It’s a meaningful shift in framing from Meta, and it’s arriving at a moment when every major AI company is racing to define what a genuinely useful “agent” looks like. Here’s what Muse actually does, what it costs, and why the way it’s being rolled out matters for markets like India.

What Muse Actually Is

Meta is calling Muse “the world’s first personal AI agent built for everyone,” which is a bold framing but not an entirely inaccurate one given what it’s built to do. Instead of answering a question and waiting for the next prompt, Muse is designed to take a goal, turn it into a plan, and keep working on it with minimal supervision — sending emails, booking travel, buying tickets, filling out forms, and even negotiating bills down on a user’s behalf.

Video Credit: Mark Zuckerberg Instagram

The product was developed under Meta’s chief AI officer, Alexandr Wang, and runs on Muse Spark 1.3, the company’s most capable model family to date. Internally, the project reportedly carried the code name “Hatch” before launch.

Users can personalize the experience by naming their agent, creating an avatar for it, and adjusting how it communicates — a deliberate design choice meant to make an otherwise complex piece of technology feel approachable rather than intimidating.

The Tech Behind It: Muse Secure VM

The part of this launch that’s drawing the most attention from security and privacy circles isn’t the task list — it’s how Meta built the infrastructure underneath it. Muse runs on something called Muse Secure VM, a dedicated, isolated virtual machine that houses both the agent and a user’s data separately from Meta’s broader systems.

According to Wang, the agent operates within its own isolated environment and “never sees your actual passwords or payment details” directly, which is meant to address an obvious concern: if you’re handing an AI agent access to your email, calendar, and payment methods, what stops it from being a security liability the moment something goes wrong. The browser Muse uses to complete tasks is also visible to the user in real time, rather than working invisibly in the background.

On the data side, Meta says interactions with Muse won’t be used to train its models unless a user actively opts out of opting out — meaning training is on by default, but Meta has said it strips “critical personally identifying information” before using conversations to improve its AI.

Where You Can Actually Use It

Muse is designed to work the way people already communicate. It’s available through its own dedicated app across iOS, Android, and muse.ai, but it also works directly inside WhatsApp, which is arguably the more significant distribution decision here. A version of Muse embedded in WhatsApp means Meta isn’t asking people to adopt a brand-new app just to try an agent — it’s meeting them inside a messaging habit they already have.

The WhatsApp version does come with a caveat: it doesn’t carry the full feature set of the standalone app, so the deepest task-execution capabilities are currently tied to using Muse directly.

Pricing: Free, Power, And Maximum

Meta is running Muse on a three-tier structure. There’s a free tier covering most everyday use, a $20-a-month “Power” tier, and a $100-a-month “Maximum” tier for people who need significantly more usage. That pricing ladder closely mirrors what OpenAI and Anthropic have already normalized for their own premium AI subscriptions — a low-cost entry point sitting beneath a much higher tier aimed at heavy users who burn through more compute.

For now, Muse is US-only and restricted to users 18 and older, which puts a hard limit on how much of this rollout is relevant to markets outside America in the immediate term.

Why The India Angle Is Worth Watching

That US-only restriction is exactly why this launch deserves a second look from an Indian market perspective rather than being filed away as “not relevant yet.” WhatsApp’s single largest user base in the world is in India, and Meta building Muse to work natively inside WhatsApp — rather than only as a standalone app — signals where the company likely intends this product to eventually scale hardest.

There’s no confirmed timeline for an India rollout yet, and Meta hasn’t said anything official about international expansion plans. But given how central WhatsApp already is to daily life for hundreds of millions of Indian users — for payments, business communication, and increasingly commerce — an agent that can act inside that same app, rather than requiring a separate download, is a very different proposition here than it is in markets where WhatsApp isn’t the default messaging layer.

A Launch That Arrived Under Scrutiny

It’s also worth noting the timing. Muse launched during what several outlets have described as a broader reckoning for the AI industry around privacy and safety, with scrutiny already building around how much access agentic AI tools should have to a person’s accounts, payments, and personal data. Meta’s heavy emphasis on the Secure VM architecture and its “never sees your passwords” messaging reads less like a marketing flourish and more like an attempt to get ahead of exactly that concern before it becomes a bigger story.

Whether that framing holds up under real-world use remains to be seen. Early hands-on coverage has generally treated the security model as the most genuinely differentiated part of the launch — more so than the task list itself, which overlaps heavily with what competing agents from OpenAI and Google are already attempting to build.

What Happens Next

Muse is still early. It’s US-only, capped by age, and split across three pricing tiers that will take time to prove out in terms of real usage. But the direction is clear: Meta is positioning itself to compete directly in the “agent” category rather than the chatbot category, and the WhatsApp integration is the detail that matters most for anyone watching this from outside the US. If and when Muse expands beyond America, India is one of the markets where its WhatsApp-first approach could matter more than almost anywhere else in the world.

FAQs