The Long Pour: Muse AI

On 8 September 2026, Meta launched what it is calling "the world's first personal AI agent built for everyone," a product named Muse that marks the company's biggest bet on consumer AI to date. The distinction Meta is drawing is between a chatbot and an agent. Where a traditional assistant answers your questions, Muse is designed to actually go and do the work, carrying out multi-step tasks on your behalf. Meta pitches it as a tool that will handle the small administrative friction of modern life, from booking movie tickets and scheduling a tennis lesson to online shopping and even filling out a permission slip for a child's school field trip. Built on the latest generation of models developed under Meta's chief AI officer Alexandr Wang, Muse was a centrepiece of Mark Zuckerberg's recent 6,500-word manifesto on the future of personal intelligence, and it rolled out first to US users aged 18 and over. The market liked what it saw, lifting Meta's stock even as the wider tech sector slid on the day of the announcement.

The way it works reveals both its ambition and its central risk. Muse lives inside a chat interface resembling a text thread, but users can name their agent, design an avatar for it, and customise how it speaks, and it is built to be far more proactive and long-running than an ordinary chatbot. Under the hood, each agent runs on a dedicated virtual machine in Meta's cloud, driving a built-in web browser that stays visible to the user so they can watch it click through websites on their behalf. To be genuinely useful, though, Muse needs access to the raw material of your life, connecting to your email, calendar, and payment methods, along with the apps you use for health, fitness, dining, shopping, music, your smart home, and more. This is the heart of the bargain Meta is asking users to strike, since to let Muse act for you, you must hand it more personal information than social media ever required, and place that trust in a company with a long and contested history over exactly this issue.

The pricing and the surrounding context are worth knowing before anyone dives in. Muse comes in three tiers, a free version that Zuckerberg says allows up to 100 million tokens a week, a "Power" tier at $20 a month, and a "Maximum" tier at $100 a month, with the paid options reportedly buying more usage headroom rather than extra features. It is a separate product from Muse Spark 1.3, a developer-focused coding model Meta released days earlier, though both run on the same underlying family. The timing is delicate, arriving less than two weeks after Meta agreed to an $18 billion multistate settlement over social media's harms, and while Zuckerberg has been repeatedly hauled before courts and Congress over how the company protects, or fails to protect, its users. Meta says it built Muse to be safe, secure, and private, and is working on a more confidential version, due before the end of the year, in which even Meta cannot see what happens inside a user's workspace. The launch pushes Meta squarely into the escalating race against OpenAI, Google, and Microsoft to control the agent that acts on your behalf. Whether people will grant a company that monetises attention such deep access to their daily lives is the real experiment now underway, and it is one worth watching with a clear and cautious eye.

Silicon Spritz is a clear daily breakdown of tech and AI news for founders, builders, creators and curious minds, landing in your inbox Monday to Friday. The Long Pour is our occasional deep dive on one company. Subscribe to Silicon Spritz to get the next edition.