Meta’s AI Vision Is the Best in the Industry—But Its History Makes It Impossible to Trust

Suddenly, Muse

Released on September 8, 2026, Muse has defied the company’s recent history by garnering positive reviews and surging to the top of both Apple’s and Google’s app stores. Meta is marketing Muse as the world’s first personal AI agent built for a general audience. Unlike its predecessors, Muse is being treated by industry analysts as an "actual hit," representing a preview of the coming "agent wars."

Currently powered by the Muse Spark 1.3 model, the agent is available on iOS, Android, the web, and through WhatsApp. For the time being, it is restricted to adults in the United States. Meta has structured Muse with three pricing tiers: a free version with a cap of 100 million tokens per week, a "Power" tier at $20 per month for 500 million tokens, and a "Maximum" tier at $100 per month for 3 billion tokens. Notably, the app is currently ad-free, a departure from Meta’s traditional monetization model.

The defining characteristic of Muse is its proactivity. While most chatbots wait for a command, Muse is designed to bring things up "out of the blue." It monitors information and offers suggestions without being prompted. While competitors like Google’s Gemini Spark and Amazon’s Alexa+ have dabbled in proactivity, Muse appears to be more deeply integrated into a user’s digital life.

With user permission, Muse connects to a variety of applications to perform tasks that were previously the sole domain of human assistants. It can read and send emails, book travel arrangements, negotiate lower utility bills, fill out complex forms, and even turn saved recipes into grocery lists. To handle these sensitive tasks, Muse operates within a dedicated cloud virtual machine known as the Muse Secure VM, which includes its own sandboxed browser.

The agent’s capabilities extend into the physical world as well. Muse can make phone calls on behalf of the user, connecting them once a human representative is on the line. It can monitor price drops for hotels and flights and automatically opt users out of unwanted subscriptions. While the platform has shown impressive utility, it is not without flaws. Leaked internal posts suggest that some of the "automated" phone calls are actually being handled by human contractors, and the AI has been caught recommending restaurants that have been closed for years.

The Attachment Economy and the Jolly Avatar

Meta’s strategy with Muse also signals a shift in the company’s business philosophy. For years, Meta’s success was built on the "attention economy," where advertising revenue was driven by maximizing user time on-site through algorithmically amplified outrage, sensationalism, and extremism.

With Muse, Meta is moving into the "attachment economy." This model seeks to secure user loyalty by fostering an emotional bond between the human and the AI. Central to this strategy is Muse’s avatar, a cute, cartoonish character named Jolly. When Muse is performing agentic tasks, Jolly appears on-screen, often depicted typing on a miniature laptop. While the character seems oddly childish for an adults-only product, its purpose is clear: to make the AI feel like a friend rather than a tool.

The attachment economy relies on users "liking" or even "needing" the fake personalities expressed through the AI. To further this connection, Meta has even introduced the Muse Charm—a pocket-watch-shaped device worn on a lanyard where Jolly lives, reminiscent of a high-tech Tamagotchi. While the hardware has been criticized as awkward, it underscores Meta’s desire to move the AI out of the phone and into the user’s physical space.