Manus China’s AI Agent in 2025:
Manus is an autonomous AI agent developed by Chinese startup Monica. It is considered the first fully integrated digital agent capable of performing tasks on behalf of users with complete autonomy, surpassing the limitations of traditional chatbots.
Manus is based on a multi-agent architecture, acting as an executive overseeing a team of specialists. When assigned a complex task, it can handle multi-step workflows that previously required various AI tools to be assembled manually. For example, Manus can sort resumes, identify relationships between different stocks, and generate reports on available properties according to desired conditions.
It is characterized by its ability to initiate tasks automatically without waiting for instructions from the user, assessing new information and dynamically adjusting its approach, making it an invisible assistant that tirelessly performs tasks until the desired results are achieved.
However, Manus has raised concerns about privacy and data security, especially given its reliance on cloud computing, which allows users to shut down their computers while the AI is working.
Manus is currently available through an invite-only web preview, and there is no official public release date yet.
Manus’s abilities:
It is a fully autonomous AI agent capable of transforming human intentions into tangible outcomes without merely providing recommendations or suggestions:
Executing Complex Tasks
It can create customized websites, from selecting the appropriate template and designing the structure, all the way to adding content and launching the final website.
Trip Planning
It can gather information about available flights, suggest the best flight options, identify suitable hotels according to budget and location, and create a schedule that includes attractions, recreational activities, and the best restaurants.
Financial Analysis
It can analyze the performance of companies’ stocks by collecting historical data, studying market reports, and making comparisons between competitors’ performance, then providing future forecasts based on data analysis models.
Custom Website Creation
It can design websites, from selecting the appropriate template, designing the structure and adding content, all the way to launching the final website.
Education
It can help teachers create customized interactive courses, prepare detailed lesson plans, create smart tests to assess student performance, and design interactive videos that explain complex concepts in a simplified manner.
Recruitment
It provides support to hiring teams by automatically sorting through hundreds of resumes, identifying the most qualified candidates based on specific criteria, and providing a comprehensive evaluation of each applicant.
The Manos team is currently testing an upgrade to Anthropic’s latest model, Claude 3.7, which is expected to significantly improve its reasoning and execution capabilities. The agent also operates asynchronously, allowing users to assign it tasks and let it run independently.
Manus is considered a significant step forward in the field of artificial intelligence, demonstrating China’s ability to develop leading AI models, reinforcing the notion that it is closing in on the United States in the AI race.
Technology and Architecture:
The Chinese AI agent, Manos, relies on an advanced technical architecture that integrates large language models and reinforcement learning models. These models operate independently within a “multi-agent” architecture based on parallel computing, where tasks are distributed among agents specializing in planning, execution, and verification. Each agent operates within an independent virtual machine and communicates via APIs or message queues. Tasks are executed in isolated test environments to ensure no interference, with support for cloud expansion to enhance efficiency and scalability.
This architecture enables Manos to break down complex instructions into executable steps and call the appropriate tools to successfully complete operations, enabling it to execute multi-step tasks that previously required human intervention or various AI tools assembled manually.
Performance and Comparison:
Manos’ performance was evaluated using the GAIA benchmark by research teams from Meta AI and Hugging Face. This benchmark aims to test the capabilities of artificial intelligence in dealing with realistic questions that require thinking, using multimedia, browsing the web, and interacting with tools.
Manus outperforms OpenAI, especially on complex tasks, demonstrating its leadership in the field of artificial intelligence agents. It is characterized by its complete independence in execution, providing ready-made final results without the need for continuous human guidance.