$META 'We see a key advantage for Meta being scale & distribution' - Goldman Sachs
With the introduction of several action-based consumer agents in the last few weeks (Spear Street Technology’s Instinct; Meta’s Muse AI; etc.), the consumer AI landscape has begun the process of undergoing a fundamental paradigm shift, transitioning rapidly from passive, conversational chatbots to proactive, autonomous agents capable of executing real-world tasks. We see the industry moving from simply answering questions to completing complex, high-value personal workflows such as travel booking, bill negotiation, calendar management, shopping, and administrative tasks.
Muse is Meta’s full scale personal AI agent app/website. We’d highlight a few key details/takeaways from the launch:
Interface and Access: Users interact with Muse through dedicated iOS and Android applications, the web (t.co/cj96ljjy2t), and WhatsApp. Meta designed the product around existing messaging behaviors to minimize the learning curve.
Capabilities: Powered by the Muse Spark 1.3 model family, which is optimized for long-horizon agentic tasks, Muse can autonomously browse the web, manage workflows, fill out forms, make purchases, and coordinate long-term goals.
Monetization: Meta has initially launched the Muse product as a paid subscription (Meta’s first major foray into paid consumer AI subscriptions). Current subscription tiers include: Free Tier (capped at 100 million weekly tokens); Power Tier ($20/month for 500 million weekly tokens and faster response times); and Maximum Tier ($100/month for 3 billion weekly tokens and priority access). We see the potential for additional monetization avenues over time, including commerce (agentic commerce capabilities with potential to charge a take rate over time) and/or advertising.
Infrastructure: Each user's agent runs inside an isolated Linux container on a dedicated virtual machine (Muse Secure VM) in the cloud, allowing it to continue executing tasks in the background after the user closes the application. Muse also has a number of partnership integrations to facilitate safe & secure interactions with various 3P networks, including most prominently Stripe (integration with Link’s wallet for agents to facilitate agentic commerce).
Security & Privacy: The transition from information retrieval to autonomous execution introduces several security, privacy, and transactional challenges. To mitigate risks, Meta utilizes its Sentinel system, a system-isolated process that acts as the sole permission authority for network access and connector actions (with credentials and authentication tokens stored outside of the agent's runtime). Meta has confirmed that Muse data/content is fully encrypted (using the same technology featured in WhatsApp) and not shared with the company (excluded from its advertising systems, with sanitized interaction data used for model training by default unless users manually opt out).