Disclaimer: This article is from the WeChat public account Phoenix Technology, authored by Liang Siqi, and authorized for republication by Stationmaster Home.
At 10 PM on March 5th, as a user with the ID "hidecloud" pressed the publish button on Bilibili, he and his team of dozens of colleagues working overtime had no idea that a small test would cause such a stir.
That night, the world's first general-purpose AI Agent, "Manus," developed by the Monica.im team, made its debut, quickly attracting attention within the AI industry. A disinterested individual posted on Moments 40 minutes later, exclaiming, "Monica's new general-purpose Agent product is absolutely amazing! They deserve to be written into history!" So excited was he that he stayed up to write an article and, at 1 AM, created a WeChat group. "When I woke up, 800 people had applied to join."
Image Source Note: Image generated by AI, image licensing provider Midjourney
"Sorry everyone, I still have to work, I can't add everyone individually."
Due to the product's current invitation-only registration, and the scarcity of invitation codes, not many people have been able to use it. According to the demonstration video, Manus is a product based on the core concept of "Mens et Manus" (mind and hand), upgrading AI from a passive responsive tool to an active collaborator that can decompose and execute complex tasks.
The night of Manus's preview video release, social media exploded with discussions. Some venture capitalists saw it as one of the most promising AI targets. However, as word spread, praise exceeding the product's capabilities began to emerge. By March 6th, criticism had formed a camp, challenging Manus's technical capabilities and suggesting excessive marketing, with many believing the praise was bought with advertising fees.
Why is the market response so divided after the birth of a new product that breaks the upper limit of AI Agent capabilities?
What is Manus?
Simply put, Manus is a general-purpose intelligent agent with independent thinking capabilities, able to plan and execute complex tasks and deliver results directly.
For example, the video demonstrates its ability to screen resumes, create travel itineraries, and analyze supplier procurement. This is thanks to its underlying "Multiple Agent Architecture," which uses planning, execution, and verification agent modules to work together, simulating the complete human "Plan-Do-Check-Act" workflow. The specific division of labor is as follows:
Planning Agent: Decomposes task objectives (e.g., analyzing stock data)
Execution Agent: Calls models to handle specific operations such as data retrieval and calculation
Verification Agent: Checks results and provides feedback for correction.
Different agents may be based on independent language models or reinforcement learning models communicating via APIs or message queues.
In the GAIA benchmark test (designed by experts including Meta's Chief AI Scientist Yann LeCun), Manus performed exceptionally well, achieving a task closure rate of 87.3%, a 67% improvement over the average 52.1% of traditional single-model architectures, particularly demonstrating stronger task decomposition and closure capabilities in scenarios such as stock analysis and real estate selection.
Based on these innovations, early adopters were amazed by Manus's intern-level practical capabilities. For example, in resume screening, it can not only unzip files and process multiple resumes simultaneously without lagging, but it can also remember user preferences and automatically generate Excel spreadsheets, completely replicating the human workflow. This stems from the team's deep understanding of B-end scenarios and the service experience accumulated during the early development of WeChat official account tools, which has given the product extremely strong scenario penetration.
According to individuals who participated in the Manus launch sharing on the morning of March 6th, Manus utilizes Claude and Qwen (with additional fine-tuning) in its model capabilities. It also boasts a cost advantage, approximately one-tenth that of DeepResearch. Currently, Manus is free to use during the testing phase.
The Mysterious Team Behind Manus: 90s Generation, "Shell Game," Serial Entrepreneurs
A key figure in the Manus development team is Xiao Hong, generally known in the industry as Xiaohong or Red. Phoenix Technology has heard Xiao Hong's name many times during its visits to the AI application sector over the past six months. Most entrepreneurs speak highly of Xiao Hong, believing him to be very innovative in AI application landing.
Xiao Hong has a wealth of entrepreneurial experience. Born in 1992, Xiao Hong graduated from Huazhong University of Science and Technology with a degree in software engineering. He has undertaken several entrepreneurial projects, including the campus social platform "Tongchuangquan" and the second-hand trading platform "Yijishi." During the boom of WeChat official accounts in 2016, he launched "Yiban Assistant" and "Weiban Assistant," which precisely addressed the operational pain points of small and medium-sized enterprises with features such as drag-and-drop layout and data tracking, serving over 2 million enterprise users before being acquired by a unicorn company, marking his first successful commercial exit.
With the explosion of large-model technology in 2022, Xiao Hong led his team to launch the browser plugin Monica in December of the same year. By integrating mainstream models such as Claude and the GPT series, it entered the overseas market with a "natural language creation tool + shared workflow" model. By the end of 2024, the plugin had over 10 million users, becoming a leading product in the overseas AI plugin sector and laying the technical and scenario-based foundation for Manus's development.
This has given the industry a new assessment of the Monica team: Taking the "shell game" to the extreme is also a kind of success. Xiao Hong often jokingly says, "Shells have their uses!"
Another core member of Manus, Zhang Tao, calls himself a classical product manager, with experience in tools, content, and SaaS. Since 2022, he has been exploring AI application scenarios. Zhang Tao was responsible for editing and creating the product demonstration video and was the first to respond after the controversy arose.
The Divided Voices
Manus's launch was a positive development for AI application landing, especially its innovative interaction method.
"Regardless, the emergence of a new product is always a good thing," an industry insider told Phoenix Technology. However, this new product launch fell into two pitfalls: First, ordinary users used descriptions similar to DeepSeek when expressing their personal feelings, and subsequent media reports did not verify these descriptions; second, some people were speculating on second-hand platforms, with invitation codes reaching as high as 50,000 yuan (now removed).
The excessive praise and descriptions led to a backlash from some industry professionals.
Even the core team did not anticipate such a strong reaction. Zhang Tao, one of the main developers, responded on social media that they would focus on building the product in the short term. He stated that they appreciate the attention Manus has received and clarified: 1. They have never opened any paid channels to obtain invitation codes; 2. They have never invested any marketing budget; 3. During the internal testing phase, system capacity is limited, and they will prioritize the core experience of existing users and gradually release invitations in an orderly manner.
As AI enters the mainstream, some new voices need to be addressed. For example, we should use AI wisely but not blindly believe in it. We should acknowledge the birth of every innovative product, but we should not overhype it, especially for startups like Manus.
"The past 17 hours have been an adventure full of unexpected events for the team," as Zhang Tao's second response stated, "Manus is still a baby in its infancy, far from the experience we want to deliver in the official version." We hope the public can be more tolerant and understanding of this small startup company with only a few dozen employees.