Meta Platforms’ new AI agent, Muse, rapidly reached the top of the U.S. Apple App Store’s free-app chart after its Sept. 8 release, helping trigger a sharp reassessment of the company’s AI opportunity on public markets. Muse reached No. 1 within 10 days and held the position for three consecutive days through Sept. 18, according to the figures provided, as users tested the assistant for tasks such as seeking refunds, identifying unused subscriptions and comparing insurance prices.
Meta shares rose 11.4% on Sept. 21, adding an estimated $192.3 billion in market value in a single session. The rally extended to AI hardware suppliers: Advanced Micro Devices gained 9.9%, lifting its market capitalization above $1 trillion for the first time.
The reaction reflects a market view that consumer AI agents may begin generating value through practical, repeatable household tasks rather than solely through chat, image generation or search. Muse’s early appeal appears to rest on saving users money or time in areas where they would otherwise have to navigate multiple websites, account portals and customer-service channels themselves.
Muse is distributed through WhatsApp and offers free computing capacity during trials, reducing the immediate cost for users who want to test agent-based software. Meta also designed the product to suggest actions based on a person’s stated goals and previous conversations. That feature followed early testing in which users faced with a large menu of possible tasks were often uncertain where to begin.
A cloud computer that keeps working
Muse operates through a cloud-hosted computer that can continue a task after a user closes the phone app. That model gives the agent more persistence than a conventional chatbot session, allowing it to monitor a page, complete multi-step forms or return to a task after waiting for a response.
Meta’s technical design separates each user into a dedicated virtual machine, or VM, while restricting individual tools inside containers. Credentials and persistent task data are managed under separate controls, according to the company’s description. A permission-checking component called Sentinel screens external access and connector actions before the agent interacts with outside services.
The company’s architecture also places model inference outside the user’s virtual machine. Instead of running the full AI model inside a personal environment, the VM communicates with external infrastructure through controlled interfaces. This structure is intended to limit the systems and data accessible from any one task environment.
The setup resembles other cloud-computer products that use sandboxed virtual machines or persistent cloud browsers. Muse’s more distinctive feature is its attempt to divide permissions, account credentials and tool execution around the boundary of a user’s personal data.
Meta can pause an idle virtual machine, save its task state to disk and resume it when more work arrives. That means a dedicated VM does not necessarily require CPU and memory capacity around the clock. Computing demand depends more directly on concurrent tasks and their duration, a calculation that will shape the cost of any large-scale free trial program.
Amazon block exposes access limits
Muse’s rapid consumer rollout has also brought it into conflict with the platforms agents are meant to navigate. Amazon blocked Muse from browsing and purchasing items on Sept. 21, citing its rules against unauthorized access and transactions.
The episode illustrates a constraint facing AI agents: a person may authorize software to act on their behalf, yet a website can still prohibit automated access under its own terms and technical controls. Login systems, fraud defenses and transaction-risk checks can interrupt an agent even when it possesses user-provided credentials.
Financial services and commerce sites pose particularly difficult conditions because they often require step-up authentication, monitor unusual behavior and tightly control account actions. An agent that can compare products, fill in forms and collect information may therefore face a different set of barriers when it attempts to move money or place an order.
Meta had already delayed Muse from an originally planned April launch to add safety measures, according to the company. Internal testing also identified reliability and privacy problems, including a ticket-monitoring task that reportedly stopped refreshing after about 15 minutes and separate cases involving exposure of private data.
Those issues place pressure on Meta to show that the product can handle long-running work reliably while preserving user control over sensitive accounts. The challenge is more demanding than producing a polished chat response: an agent must maintain context, recognize when it lacks authority and avoid making an irreversible mistake while carrying out a task over time.
Revenue forecasts depend on paid retention
Sell-side forecasts have attached substantial potential revenue to a successful paid version of Muse, though those estimates depend on adoption and computing costs that remain uncertain. Truist projected that Muse could add $28.5 billion in incremental revenue to Meta by 2030.
Oppenheimer estimated that roughly 115 million paying users would be needed to materially affect Meta’s profitability, based on a $20 monthly subscription price and an 80% incremental operating margin. Its model also indicates that generous free computing allocations could reduce margins, especially if users run lengthy or resource-intensive tasks.
Meta shares later fell 0.63% on Sept. 22 before rising 0.6% in after-hours trading in the period referenced. The next test is likely to be retention after the initial trial phase: whether users continue assigning Muse recurring tasks once the novelty of an autonomous assistant fades and whether those tasks prove valuable enough to support a paid service.
The claims surrounding Muse do not yet establish a direct investment case for cryptocurrency networks, decentralized computing tokens or machine-to-machine payment protocols. Agent platforms still depend heavily on access to established websites, identity systems and payment rails, and Amazon’s restriction shows that open technical standards alone would not remove those commercial and policy barriers.
Explore how AI reshapes finance and trading with Toobit’s AI and blockchain insights for data-driven market strategies.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
