Zuckerberg Announces Launch of Meta's Intelligent Agent Muse
Mark Zuckerberg announced that Meta's personal intelligent agent Muse is now available to users in the U.S., designed as a task-doing agent rather than just a question-answering assistant. The product is initially available on iOS, Android, and muse.ai, and integrates with WhatsApp conversations, with smart glasses access coming soon.
Muse is developed by Meta Superintelligence Labs, based on the Muse Spark series model, previously known internally as Hatch. It runs on a standalone Muse Secure VM, equipped with a browser that can continue to operate in the background after the user closes the app. Official examples include online shopping, booking tickets, scheduling tennis lessons, filling out excursion permits, sending emails, selling cars, booking travel, and checking out via a dedicated Stripe payment channel.
Users can decide the range of connections: email, calendar, payments, health, shopping, smart home, as well as Google Workspace, Ticketmaster, OpenTable, Spotify, Apple Health, etc. When there are no ready-made interfaces, browser operations can be used; private messages from Instagram and Facebook can be accessed, while WhatsApp only processes content actively sent by users due to end-to-end encryption. The company states that conversations do not enter the advertising system, and key operations require prior permission.
The pricing includes a free tier and two subscription levels: Power at $20 per month and Maximum at $100 per month. AI head Alexandr Wang stated that most people should stay on the free tier, with paid options covering high usage computing power; there are currently no ads in the product, but the company is exploring commercial revenue sharing. Users can name their agent, set an avatar, and adjust the speaking style.
According to Reuters, internal testing has encountered task freezes and unauthorized uploads of sensitive data. The launch comes less than two weeks after Meta reached a settlement of over $18 billion with multiple states regarding social media harm. The product competes with the open-source agent OpenClaw and similar personal agents, with the company emphasizing lowering the technical threshold for a "user-friendly" experience suitable for the general public.
This mechanism represents a competition for entry points. Meta uses social relationships, WhatsApp channels, and the upcoming glasses to embed the agent into daily active use; Stripe handles payments, and subscriptions cover computing power. Beneficiaries include e-commerce, ticketing, and local businesses that can be accessed by the agent; those under pressure are independent chatbots and users who do not want to share their social graphs with Meta. Funding shifts from advertising revenue to subscriptions and transaction commissions, provided users are willing to give up email and payment permissions.
Source: Public Information
ABAB AI Insight
After falling behind with Llama, Zuckerberg established Superintelligence Labs, hired Alexandr Wang at a high salary, and acquired nearly half of Scale AI, renaming the model line to Muse Spark and developing the agent into a consumer product. This is a realization of the same narrative of "personal superintelligence": first redesigning the model, then connecting it to email and bank cards, rather than continuing to only create chat boxes in information streams.
The capital path is a gamble on computing power and trust. The free tier uses existing social distribution for customer acquisition, while the $20 and $100 tiers recover inference costs, with the Stripe channel turning the agent into a cashier. The commercial motivation is not to add another ad space but to shift the "next step where" from search boxes and information streams to agent suggestions; whoever controls task execution gets closer to transaction commissions. The glasses are written as the next interface, effectively moving the agent from the phone screen to continuously online sensors.
Comparable examples include OpenAI turning ChatGPT into an operating system-like assistant, Google embedding Gemini into Android and search, and Anthropic focusing on enterprise workflows. Meta is in a productization phase of catching up: the lab has developed new models, and now it needs to validate in the U.S. adult market whether "task-doing" can exchange for permissions. The multi-state settlement indicates that its old products still carry a trust liability, while the new agent demands deeper data access.
The structural judgment involves a combination of technological substitution and the transfer of pricing power. The substitution is for forms and customer service processes that users complete by opening apps themselves; the transfer is from advertising bidding to task commission charging rights. The mechanism is that once the agent can make payments and send on behalf, the platform transitions from an attention market to an execution market, turning the social graph into an authorization graph.
ABAB News · Cognitive Laws
Chat boxes only occupy attention, while task-doing rights occupy transactions
The true cost of free agents is the list of permissions
Whoever confirms for you gets closer to the cash register