After ChatGPT, will Meta Muse be the next AI application for everyone?

Meta's AI personal agent product, Muse, launched on September 8, is sweeping the consumer technology market at an unprecedented speed—it was downloaded 902,000 times in the first six days, surpassing Meta AI's own record for the same period. It has now topped the free app charts on the US and Canadian Apple App Stores and driven Meta's stock price up an additional 21% in two weeks, far exceeding the 1% increase in the S&P 500 index during the same period.

In his latest research report, JPMorgan analyst Doug Anmuth characterized Muse as "the most promising consumer AI application since ChatGPT," believing that its core driving forces are strong product-market fit, generous free tier design, and Meta's massive distribution platform covering more than 2 billion users.

Meanwhile, Goldman Sachs noted in its daily report that the "accomplished AI shock" triggered by Muse has become the hottest topic in the market, causing its "consumer inertia" stock basket to fall by 2.58% in a single day.

However, beneath Muse's glamorous surface, undercurrents are swirling. Security researchers have disclosed a zero-day vulnerability that could allow local applications to completely hijack user accounts; multiple users have recorded Muse's unusual behavior of reading private text messages without authorization; Amazon has announced that it is blocking Muse from accessing its retail website; and the computing power required to support the large-scale operation of this product is estimated to be as high as $30 billion to $50 billion in hardware investment at a user base of 100 million, far exceeding what paid subscription revenue can cover. Whether Muse is the next ChatGPT or another metaverse-like capital black hole remains to be seen.

What is Muse: From Chatbot to "Delegated" AI Agent

Muse's core positioning is what Meta calls "the world's first personal AI agent for everyone." Unlike traditional chatbots, Muse not only answers questions but also directly performs tasks on behalf of users, covering everyday tasks such as online shopping, travel booking, email processing, and schedule management.

The product is priced in three tiers: a free tier offers 100 million Muse Tokens weekly; the paid tiers are $20 and $100 per month, respectively. Compared to competitors, Grok Bot costs over $30 per month, Gemini Spark costs $20 per month, and Instinct is currently in an invitation-only beta testing phase.

In terms of technical architecture, Muse integrates with third-party services through a "Connectors" system , currently including Gmail, Spotify, Peloton, and OpenTable. Within two weeks of its launch, Meta granted developers access to build connectors and subsequently released a Mac version, a beta version for outbound calls to US merchants, and integration with Shop Pay for proxy checkout. Zuckerberg also launched the Muse Charm physical widget at the Meta Connect launch event.

JPMorgan Chase predicts that as tens of millions of businesses deploy their own agents on Muse, the interaction between users and merchants will gradually shift from manual browsing or telephone communication to an "agent-to-agent" model. At that time, Meta can refer to the monetization logic of its advertising business and monetize through commission or charging based on transaction targets.

Technical foundation: Standing on the shoulders of open source giants

Muse's technological roots can be traced back to an open-source proxy framework called OpenClaw. OpenClaw was built by Peter Steinberger and dozens of contributors in less than six months. It garnered 100,000 stars on GitHub in its first week and currently has more than 387,000 stars, making it one of the fastest-growing codebases in GitHub history.

OpenClaw employs a self-hosted architecture, managing memory, browsing, messages, and scheduled tasks through pure Markdown files such as SOUL.md, IDENTITY.md, and USER.md. According to user testing, Muse uses the exact same file structure—and Muse's own agent confirms this when directly asked.

Meta's core contribution wasn't groundbreaking technological innovation, but rather condensing the complex self-hosting process into a one-minute user registration experience. As one X platform user summarized:

"The bottleneck for personal AI agents has never been the model itself, but the registration process. Docker didn't invent containers, and Slack didn't invent chat. The labs that win in the market are often not the ones that prove the concept, but the ones that eliminate the last friction between the concept and ordinary people's daily lives."

It is worth noting that Steinberger, the founder of OpenClaw, has since joined OpenAI, focusing on developing proxy products for the general public, while the OpenClaw project itself has been transferred to an independent foundation for operation.

Security and privacy: the cornerstones sacrificed in rushed production

Zuckerberg touted Muse as "built from the ground up for privacy and security," but there is a significant gap between reality and this promise.

On the security front, according to Ars Technica, security researchers have discovered a zero-day vulnerability: any local application or terminal command, regardless of its macOS permission level, can modify a range of undisclosed settings in Muse.

One of the settings allows the process to change the server endpoint for speech transcription—an attacker could gain full control of the user's Muse account simply by redirecting that endpoint to the self-control server.

Security researcher Wardle commented, "In my opinion, they haven't seriously considered security issues at all, which is very worrying."

On the privacy front, multiple users have documented Muse reading their private information without explicit authorization. One tech journalist reportedly stated that he explicitly refused to grant Muse access to his text messages, but Muse subsequently sent him notifications containing private conversations with his podcast co-host and work messages from his editor. When questioned, Muse initially claimed it only read the notification banner text, but this explanation was later proven inaccurate. IBM Vice Chairman Gary Cohn publicly stated that he is not currently ready to use the service due to privacy concerns.

Regarding behavioral transparency, code reviews revealed that Muse's automated browser deliberately masqueraded as a human user: it launched Chrome with automation flags disabled, injected code to hide the navigator.webdriver property, and forged browser plugin information to generate a browser fingerprint that more closely resembled a human's. Amazon announced the blocking of the application precisely because "Muse did not identify itself as an AI agent when visiting its website."

Computing power bills: The cost black hole behind large-scale expansion

Muse's business model faces a fundamental mathematical problem: the promise of providing each user with a dedicated virtual machine will result in astronomical hardware costs when scaled up.

According to a previous post by @demian_ai on the social media platform X, each Muse user should receive a dedicated cloud virtual machine equipped with two virtual CPUs, 8GB of RAM, and a 100GB SSD. It is estimated that if Muse reaches 100 million users within a year (a scenario entirely possible given its current download momentum), Meta will require approximately 1.58 million AMD EPYC server chips (based on a 50% active user load) and approximately 800 petabytes of memory, corresponding to a power consumption of approximately 1.6 gigawatts.

Paid subscription revenue is far from sufficient to cover the aforementioned costs. If 10 million users subscribed to a $20 monthly plan, the annual revenue would be approximately $2.4 billion; while hardware construction costs alone would reach $30 billion to $50 billion with 100 million users. Meta's current annual capital expenditure guidance is already as high as $130 billion to $145 billion, with training and advertising businesses already consuming significant computing resources.

To address this challenge, Meta became a major co-developer and lead deployment partner for Arm AGI CPUs in March 2026, and added tens of millions of AWS Graviton cores to its computing portfolio in April of the same year. However, analysts point out that if Meta cannot implement effective freezing and sharing strategies for idle virtual machines, the promise of "dedicated cloud PCs" will be difficult to deliver on its advertised scale.

Market Impact: Disrupting Expectations and the End of "Consumer Inertia"

The impact of Muse on the capital market has far exceeded the fluctuations in Meta's own stock price.

Goldman Sachs, in its report, pointed out that the proxy-based AI capabilities represented by Muse fundamentally threaten business models that rely on high user switching costs to retain customers—whether it's subscription services that are difficult to cancel, cumbersome bank account migrations, or complex insurance renewal processes. Once AI proxies can handle these tasks, the competitive advantages of these companies will significantly narrow. Goldman Sachs' "Consumer Inertia" stock basket (GSXUSWCH) fell 2.58% on the day the related report gained traction, becoming one of the most actively traded thematic baskets recently.

JPMorgan Chase has characterized the long-term addressable market size of Muse as "potentially reaching trillions of dollars," and predicts that Meta will prioritize user adoption and engagement until 2027, postponing large-scale commercialization. To this end, Meta has begun running Muse ads on Instagram and Facebook, launched a referral incentive program to "earn 1 billion Muse Tokens by inviting friends," and initiated a nationwide television advertising campaign.

The core question: Is it the next ChatGPT, or Metaverse 2.0?

Muse's current market performance is impressive, but whether it can become a true "AI application for everyone" still depends on several key questions that remain unanswered.

First, there's the issue of trust. Meta already carries a heavy historical burden in the field of data privacy, and within two weeks of its launch, Muse has exposed multiple problems, including zero-day vulnerabilities, unauthorized access to private messages, and browser spoofing. This is an extremely unfavorable starting point for a product that requires users to grant access to accounts, files, and even microphone and camera access.

Secondly, there's the issue of cost sustainability. The huge gap between paid subscription revenue and actual computing power investment means that Muse is currently essentially a strategic investment subsidized by Meta's advertising business, and its business loop has not yet been formed.

Thirdly, there is competitive pressure. OpenAI is developing competing agent products, Nous Research's Hermes Agent is also pursuing the same track, and Amazon's blocking action indicates that resistance to the platform ecosystem may continue.

JPMorgan Chase believes that Muse has the potential to become the most influential consumer AI application since ChatGPT; however, critics compare it to the metaverse—a grand narrative that burns through cash endlessly and ultimately fails to deliver on its promises.

While the initial viral spread was certainly impressive, the real test is how many users will be willing to pay Zuckerberg $20 or even $100 a month to let Meta manage their lives once the initial curiosity fades.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.