Join our daily and weekly newsletters for the latest updates and the exclusive content on AI coverage. Learn more
Stop me if you’ve heard this one: a little-known Chinese startup makes waves worldwide for an impressive new AI product.
No, we are not talking about Deepseek R1, the AI reasoning model that made waves among Western IA circles earlier this year. Instead, the new hot product of the day is called ManusAnd it is a new versatile IA agent-that is to say, more than a model of AI, it is an interface to control several models which can go out and, in itself, independently of complex tasks such as generating reports or performing dozens of social media accounts in the name of the user.
If it looks like in -depth research modes offered by Openai, Google and others, as well as the Openai operator agent and the mode of use of the Anthropic computer (the last two of which can, as a manus, take control of the computer or a user’s programs, you have understood what it aims to offer.
But for leaders and decision -makers focused on companies here in the West and abroad – such as technology leaders, product managers, computer team leaders, and more – what should they know about manus and the capacities it offers? Read the rest to discover it.
What is Manus and who is behind?
Manus ai was officially Announced on March 5, 2025 on the social network X With a position of the company describing it as “the first agent of AI general who does independently performs complex tasks rather than simply generating ideas.
According to a South China Morning Post report (SCMP)Manus was developed by a Chinese company called Butterfly Effect, which has offices in Beijing and Wuhan. The company would only have a few dozen employees, but quickly attracted attention to the landscape of Chinese AI.
The founding team includes experienced entrepreneurs and product managers, led by Xiao Hong, a 33 -year -old serial entrepreneur and graduated in 2015 from the University and Huazhong technology from Wuhan.
Xiao previously built applications based on WeChat which were acquired by large companies and then launched Monica.ai, an AI assistant available as a browser and mobile application extension.
On his website, manus would be appointed for the Latin word for “hand”, a nod to the fact that users can count on him to carry out actions for them, or, in my words, to “lend them a hand”.
How does Manus Ai work?
Manus AI is designed as a multi-agent system, which means it combines several AI models to manage tasks independently.
Unlike AI chatbots that help users by providing information, manus can search, analyze data, generate reports, automate workflows and even write and deploy code.
According to X messages from Ji Yichao, co -founder and chief scientist of Manus AI, the system is built on the Sonnet Claude 3.5 of Anthropic – a model of nine months at this stage – and refined versions of the Qwen models of Alibaba.
The team is currently testing the manusing upgrade to the most recent and efficient anthropic model, Claude 3.7, which should further improve its reasoning and execution capacities.
Manus AI works asynchronously, which means that users can assign tasks and move away while it completes them independently. It is currently in a private beta version, with access granted by invitation codes.
How is Manus AI accompanied by competition based in the United States?
One of the biggest reasons why AI has gained ground is its strong reference performance, beating the in -depth research agent of the American company Openai and “the” anterior point “, according to a graphic published on the official site of the manus.
This assertion, as well as the real world tests, led some users of EIA energy and the first adopters to the conclusion that manus can be one of the most competent AIA agents available today.
Beyond the references, Manus has already proven itself on independent platforms such as Upwork and Fiverr and in the automatic learning challenges of Kaggle, successfully performing complex tasks of the real world.
AI influencers celebrate the arrival of manus and impressive performance
The conversation on manus in the media and circles of AI took off at the end of last week when users on social network X noted that some people used it to automate the management of up to 50 corporate accounts at the same timeIn real time, showing its capacities to create commitment fleets that companies could use for opinions.
In addition, although this has not yet been proven for manus, the same technology could probably be used for all kinds of marketing and influence campaigns, even political propaganda or disinformation.
But for the most part, users and influencers of the EIA from AI in the West have been largely impressed and celebrated the arrival of manus – impressed by their first tests once they received some of the rare beta invitations, or observed the work of others with access to the tool.
Rowan Cheung, founder of The Newsletter Rundown Aidescribed the launch of manus ai as a potential turning point for AI agents and said that “the second deep moment in depth of China is here” A message on his LinkedIn account.
“This AI agent called” manus “is going crazy in China at the moment … It’s like a deep research + the operator + Claude Computer Combined, and it’s really good.”
Cheung personally tested the manus and discovered that he:
- Created and deployed a biography website on itself, with 100% precision and real-time data recovery.
- Have found the best rental spots in San Francisco according to the crime rates, the presence of the AI industry and the density of entrepreneurship.
- Has developed a complete AI lesson, generating eight content chapters, including tools, use cases and prompts.
He received 500 invitation codes from the manus team and deposited them to his subscribers and readers.
Youtuber focused on Googler and Ai Bilawal Sidhu shared a practical video review, calling manus “the closest thing I saw an autonomous AI agent”.
“It’s as if you are stuck above someone’s shoulder who uses a computer … asking what to do at the highest level, and that does it essentially for you.”
Sidhu tested the manuspot on various tasks, including:
- Search for places – Scanning Google Maps and Information Sources to recommend the best places according to regulations, accessibility and security.
- Development of video applications – Automating of video effects such as ASCII art and real -time filters.
- Extract information from the industry – drawing data from Reddit, Twitter and other sources to generate comprehensive reports.
Sidhu stressed that Manus decomposes complex work flows and performs them step by step, making it one of the most independent AI agents available today.
Another influencer from AI, Chubby (@kimmonismus on x)said Manus Ai has outraged in -depth research of Openai: “I am not paid for the promotion. I just tried access to the overview, and he outperformed the in -depth research of OpenAi a lot! Breathtaking. “”
Andrew Wilkinson, co-founder of the startup exit supplier and Tiny buyer, Written on x It’s “it’s absolutely crazy. I have the impression of traveling in time six months in the future. I launched a zip file of 20 candidates for a CEO work and he made a deep dive on each, one by one… ”
Large questions about the future of manus
Despite the excitement, questions and concerns have emerged concerning the real capacities of AI.
According to the South China Morning Post, very few people were able to test MANUS IA due to serious shortages of servers after its sudden reputation.
Zhang Tao, a product partner of Manus AI, admitted to a publication on social networks: “The current mechanism of the invitation only is due to a server capacity truly limited to this stage … The team has underestimated the enthusiasm of the public response, and our server resources were only provided for a level of demonstration.”
This lack of availability has fueled speculation that Manus AI could use rarity marketing tactics to generate media threshing. The Invitation Codes for Manus have been sold on the China market, Xianyu, leading to frustration among potential users.
In addition, users such as @Teortaxestex on X (the “Stan # 1”, or obsessive, Deepseek), estimated that “it is a diabolically optimized product for influencers, which is why it has exploded so much. Generate threadboy content, travel plans and such a general interest [mind blown emoji, hand pointing down emoji] Tips – Yah. STEM assistance, coding – Worse than Google. More LLM than the agent. »»
In other words, some believe that it is good for influencers who seek to build a follow -up of the spam lists and advice advice, but not as good as other tools at work such as development. Even the former Venturebeat and current writer Techcrunch-Er Kyle Wiggers Said in an article during the weekend, “Manus is probably not the second” deeply “moment of China.”
Is Manus Ai the real deal?
Unlike Deepseek R1, which has developed its own foundation model, AI manus is based on existing large language (LLM) models, such as Claude and Qwen. This has led some criticisms to wonder how original manus technology is.
Yichao Ji recognized it in the official introduction video, declaring: “Manus works as a multi-agent system powered by several distinct models.”
Despite this, Ji stressed that Manus AI is committed to developing open-source and will publish some of his models to the public later this year.
Currently in private beta, AI manus is only available via invitation codes. The company plans to extend access soon, but the server limitations can delay a full public launch. An open source component is provided, allowing developers to integrate parts of its AI system.
While some consider manus as a breakthrough in autonomous AI agents, others remain skeptical about its true capacities, commercial model and scalability. While the media threshing around manus has continues to grow, the question remains: is this the next moment deeply, or just another over-type AI product?
