Meta Muse Pioneers the Personal AI Era

Meta's stock price surged 11% in one day following the app's release, hitting a record high of $777. The AI agent Muse, launched on September 8, can be used immediately upon login without any setup process, and it surpassed ChatGPT to become the #1 free app in the U.S. App Store within ten days.

This achievement propelled Meta's stock price up 11% in a single day, with the stock reaching a high of $777. Muse offers high convenience, being usable immediately upon login without any separate configuration.

Muse's 'Feed' feature analyzes personal information such as user location, hobbies, and family relationships to provide tailored news. For example, it automatically notifies users of hurricane forecasts when planning a trip to Hawaii.

'Idea' helps active utilization by suggesting prompts and automation tasks, like packing for a trip, based on the user's past records.

The 'Goal' feature automatically tracks and manages progress for tasks set by the user, supporting goal achievement.

Muse's 'Library' feature generates personalized web applications by writing code, such as HTML, according to user requirements. This demonstrates the agent's characteristic of not just conversing but performing actual tasks and producing results.

For example, it can automate tasks like generating YouTube content pipeline sheets, increasing user productivity.

In this way, Muse is ushering in an era of developing applications exclusively for users.

Muse performs all tasks independently on a dedicated virtual machine (VM). This VM can handle various functions including coding, browsing, file processing, and memory management.

Unlike existing agents that required complex setup procedures, Muse uses an already configured environment, providing high convenience to users.

Muse functions as a practical assistant that sets goals and independently produces results. Unlike ChatGPT, which primarily focuses on chatting, it can directly perform proactive actions such as purchasing items, negotiating prices, and scheduling appointments. Muse can directly purchase items, negotiate prices, and adjust schedules, handling concrete tasks.

When a user gives it a goal, it doesn't just provide information but also produces results, offering practical assistant functions.

Muse separates the agent's work area from the credential area through a robust security architecture. Before an agent executes, a separate agent called 'Sentinel' first reviews the safety of the request.

It provides automatic reservation functions through integration with external apps like OpenTable, and the more connectors there are, the stronger the platform's competitiveness and usability becomes.

Muse offers various pricing plans based on user needs, including Free Plan, Power Plan, and Maximum Plan. Currently, it provides 100 million tokens per week for free.

Users can set detailed Custom Permissions when linking various external services like mail, calendar, SNS, and health data, such as blocking only specific folders.

Muse can analyze emails and calendars to extract tasks and manage schedules for the user. For example, it demonstrated a feature that finds tasks based on the past 7 days of emails and appointments.

When planning a trip, it compares flight prices from San Francisco to Incheon and recommends three accommodation options near Nonhyeon-dong, Seoul, including total cost and cancellation conditions.

It also demonstrated functions for recommending baby product shopping items and preparing for payment, providing various automation features tailored to the user's daily tasks and personal needs.

Muse demonstrated a shopping automation feature, using Aveeno lotion as an example, comparing items, quantities, and prices, and preparing for payment right up to the final step. Connecting an Amazon account allows for even more personalized shopping.

For moving from San Francisco, it also showed a feature comparing three local moving companies, primarily Korean-owned, based on budget, date, and house size. By inputting information, Muse can even make proxy calls to confirm quotes.

Muse executes all tasks on a dedicated cloud virtual machine (VM), not on the user's local computer. Therefore, even if the user closes the application, the task continues to run in the cloud.

The cloud environment includes a browser, files, and code execution tools, and for privacy protection, the model responsible for permissions is separated from the model that performs the actual tasks, enhancing security.

A technology developed by Meta called 'Harness' handles this loop.

For user convenience, the complex internal execution process is hidden, providing result-oriented services. Unlike existing open-source agents, it features easy setup that can be used immediately without user intervention.

Muse is expected to generate revenue by securing commissions from transactions like shopping created by the agent. Providing 100 million tokens per week for free appears to be a user acquisition strategy.

The strategy aims to preempt the market entry point through user-centric browsing and AI agent interfaces, and following its launch in the U.S. and Canada, a release in Korea is also planned.

In the AI era, companies that preempt the 'entry point' of user interfaces are expected to be market winners. Just as Microsoft dominated with OS and Google with search engines in the past, AI agents are analyzed to be the new entry point of this era.

As users shift to consuming information and performing tasks through agents, Meta intends to secure leadership in the AI era's user interface through Muse.