GuShiio.com 
(Photo source: Manus official website)
Following DeepSeek, an AI Agent agent application developed in China has attracted attention.
Domestic AI technology team Monica recently released Manus, the world’s first general-purpose AI Agent agent product.
Different from traditional AI assistants, Manus takes Mens et Manus as its core concept. It can solve various complex and changeable tasks. It not only provides suggestions or answers, but also directly delivers complete task results, using dialogue to screen resumes, research real estate, analyze stocks, etc., and achieved SOTA scores in the GAIA benchmark test, attracting attention. In the past day, the founder of #Manus has posted multiple entries such as the post-90s generation in ChinaWeiboHot searches.
However, since the product currently uses the invitation system to register and use, and the invitation code is difficult to find, more than 4500 people have joined the official discord to seek the invitation code (speaking rights have been closed at present). At the same time, on hand trading platforms such as fish,The Manus internal beta code has been fired to 100,000 yuan each.
Also,Manus has also triggered many chaos such as various QR codes to pull WeChat groups, live broadcasts, multiple investors looking for his team, and stock prices in the capital market. Data shows that China and Singapore (where investors gather) were the regions with the most Manus entries searched yesterday.
Not only that, the functions demonstrated by Manus are not significantly different from OpenAI’s Operator, Smart Spectrum AI’s Phone Use, and Cognition Labs’s Devin. Therefore, some developers question whether Manus is just a localized shell for overseas technology. Is it a hype?
On the afternoon of March 6, Manus product partner Zhang Tao @hidecloud said that the team completely underestimated everyone’s enthusiasm. The original intention was to share the phased gains in the process of exploring the form of the agent product. The server was only prepared at the demo status level and never expected it to cause such a big wave. The invitation code mechanism is currently adopted because the server capacity is really limited at the moment, so it is a last resort.& ldquo; The Manus you see so far is still a baby in an infant, which is far from the experience we want to deliver to you in the official version.” rdquo;
same day,sound networkYao Guanghua, head of the AI RTE product line, told GuShiio.comAGI and others that Manus and conversational AI products do not belong to the same species, Manus is a tool.At present, our understanding of AI Agents is still tools. The value of replacing many of your existing tool types is not the ultimate goal. Conversational AI interaction can generate emotional value. It may become an AI product that transcends tools and is similar to companionship.
An AI team that understands communication
The company behind Manus’s AI Agent product, BUTTERFLY EFFECT, has its core product is Monica.im. The main body of the domestic affiliated company is Beijing Butterfly Effect Technology Co., Ltd., which was established on April 20, 2022. At that time, both shareholders and legal representatives were 33-year-old China entrepreneur Xiao Hong.
On March 10, 2023, the company’s shareholder and legal representative were changed to Feng Qionghua. Two months later, Xue Fengying became a new shareholder, and her registered capital increased from 1 million yuan to 1.25133 million yuan. It is reported that Butterfly Effect completed Series A financing in the autumn of 2024 and has received investment from Tencent, Zhenge Fund and other institutions.
The founding team of Butterfly Effect includes serial entrepreneurs such as Xiao Hong and Ji Yichao. Among them, Xiao Hong is the founder of Butterfly Effect. He graduated from Huazhong University of Science and Technology and founded Nightingale Technology. He focuses on the application development and promotion of the WeChat ecosystem. He has received hundreds of millions of yuan in strategic financing from Tencent and other companies. His products include One-Partner Assistant, Micro-Partner Assistant, etc.
Ji Yichao is the co-founder and chief scientist of Butterfly Effect. He graduated from Beijing University of Information Science and Technology and received angel investment from Zhenge Fund and Sequoia China for developing the Mammoth Browser; Zhang Tao, another core member of Manus, is exploring AI application scenarios after 2022. He was the person in charge of Guangnian Beyond products founded by Wang Huiwen, and later Meituan acquired the equity of Guangnian Beyond.
If you review Manus, which has been on fire for 24 hours, you need to start with the product Monica.im (hereinafter referred to as Monica).
Monica is a browser plug-in that was released in early 2023 and provides services by integrating mainstream models such as Claude and GPT series.
But what many people don’t understand is that Monica is actually an AI product acquired by Xiao Hong’s team. It comes from chart for Google, an independent developer, with a M & A transaction volume of millions of yuan. Through this tool for testing and optimizing the diversion, Monica survived the cold start period,Quickly complete user accumulation.
At the same time, Monica offers a paid model,Minimum of 3 per month, minimum of 9 per year, maximum of 9 (9) per year for Pro+ versions,Enable the team to achieve revenue.As of now, Monica has more than 3 million downloads on the Chrome browser plug-in, with more than 23,000 ratings, and the total payment may reach tens of millions to hundreds of millions of dollars.
A product manager concluded that Monica’s core is to copy Jasper (the originator of GPT shells and once a phenomenal AI application), and its product form is the web. However, Monica uses browser plug-ins that are low-cost (easy-to-use, flexible) and user experience. Good (timely and convenient) and other methods, and has developed from a single application to an AI family barrel, covering more scenarios and better serving user needs.
Of course, the Monica team is not foolishly waiting for users and improving product competitiveness, but understands payment and marketing routines, and conducts marketing through overseas self-media delivery channels.
Monica operating partner Yang Huijie once revealed at an event that in addition to gadgets,Another thing the team did in the early days was to launch KOL opinion leaders on the X (formerly Twitter) platform. Usually, KOL in AI-related fields quoted US$50 to US$150 each, becauseTherefore, the team found dozens of bloggers to publish Monica-related content, which can bring millions of exposure when the results are good.
“There are three forms of tweets: list, which means that multiple tools are introduced in a tweet, which often has a large spread and a low conversion rate;tweet, which means that only one tool is introduced in a single tweet, and the spread level is small, but the conversion rate is relatively good;thread, which means that multiple tools are introduced one by one in the form of a thread, and each tool will have a more detailed introduction, with the highest conversion rate and a good spread level. Twitter was chosen as an early promotion platform because Twitter has a large number of active PC users, which fits our form as a PC product. In addition, the url in tweets supports direct jumps, so that we can use official website links with parameters on the PC to track the number of user registrations brought by each tweet. rdquo; Yang Huijie said.
Yang Huijie also said that“we observedForeign countries guide users to pay during the registration process,Generally speaking, users have not started using it yet, so it would be a bit radical to directly add paid guidance in the registration process, but this is the practice of many foreign products.The same conclusion was also reached in China. Domestic plug-ins require users to pay during the registration process.”
Yang Huijie pointed out that the reason why Monica launched the mobile app was also to broaden the channels for customer acquisition, including cooperation in making member gift packages and quickly producing a series of advertising pictures.The pre-advertising cost of such ideas may be only US$50-US$100, and there is no research and development cost.
One detail that GuShiio.comAGI noticed was that in the Monica domestic version of the App, there is a special invitation channel that allows big KOL to apply for exclusive invitation codes for content creators. This is enough to see that the team attaches great importance to KOL’s social marketing.
Obviously, Xiao Hong, Ji Yichao and others are an AI product team that knows how to market and operate through overseas active self-media accounts. An article worth just a few tens of dollars allows the public to download apps and find invitation codes like pyramid schemes.
Today, the team officially released Manus, an AI agent product, which quickly attracted the attention of all parties and attracted widespread attention.Although Zhang Tao emphasized that the company has never invested any marketing budget, it is undeniable that these funded AI self-media accounts contributed to this round of Manus dissemination, and they swarmed to publish articles at almost the same time.Promote it as the world’s first universal Agent, with strength that surpasses OpenAI, and even use expressions such as amazement, going down in history, and a symbol of national scientific and technological strength.
Furthermore, these AI self-media practices are no different from Li Yizhou ‘s 49-yuan course that sends patriotic videos and sells emotional value.
From a product perspective,Manus ‘promotional video includes three cases: one is to upload a compressed package containing 10 candidate resumes to Manus and ask Manus to help screen resumes. Manus reads all resumes after analyzing the instructions, then provides ranking suggestions, and can generate a form file at the user’s request, and its execution process is displayed through a visual graphical user interface; the other is to find real estate targets, and Manus can screen those that meet the needs according to user needs; The third is to analyze stocks. Manus can open Python to analyze and generate HTML web files.
GuShiio.comAGI also reproduces OpenAI’s future roadmap information. Manus generates code and collects data to ultimately generate visual charts and documents.
Therefore, to summarize, Manus’s core architecture is highly similar to Anthropic’s ComputerUse, relying on a multi-agent virtual machine environment to complete tasks. The so-called autonomous planning capability is actually based on existing models and API calls, superimposed on Internet data, and used Python scripts to generate visual charts. The essence is still the splicing of tool chains, not a breakthrough in the underlying technology.
The end result is that, completely different from DeepSeek, the current discussion about Manus is basically limited to the Chinese Internet. Although its product release is in English rather than Chinese, the English-speaking AI community has basically paid no attention to this. CNN, Reuters, etc. are all briefly introduced together with the Ali model, and no major media such as Bloomberg and the Wall Street Journal have followed up.
This shows from the side thatThe AI self-media previously fed by the Monica team has played a huge role in the current Manus traffic boom.At the same time, this also explains why domestic AI teams such as Manus and DeepSeek are not on the Weixin Official Accounts, but choose to publish important product articles on the X platform.
However, from the perspective of user experience, Manus is still within the scope of current AI capabilities and is not a transformative product, but the overall Agent experience exceeds market expectations.
“The past 17 hours have been tantamount to an adventure full of all kinds of surprises for the team. As Li Tao said in his second response, Manus is still a baby in his infancy and hopes that the public can give this small team of only a few dozen people more tolerance and understanding.
As of the release of GuShiio.comAGI, the Discord chat channels announced by Manus official website were all closed, only to release new cases on the X platform.
“We apologize for the inconvenience caused and deeply thank you for your continued support and understanding. Thank you for being a member of our community!& rdquo;Manus official announcement stated.
In 2026, more than 80% of enterprises will integrate AI agents into production processes
Market analysts said that the significance of Manus’s ecological strategy is to introduce a new business model for the AI industry to build a common AI agent ecosystem, which may become the second largest AI application scenario after AI search.
Although the Manus craze is only a flash in the pan, the development trend of AI Agent agents deserves long-term attention.
A report released by Gartner in January last year showed thatCurrently, 21% of companies have integrated AI agents into their production workflows, and by 2026, this number is expected to exceed 80%. In addition, Gartner predicts that by 2028, 15% of daily work decisions will be made by agents.
Today, AI agents have demonstrated advantages in many fields such as medical care, e-commerce, and network security, helping enterprises reduce costs and increase efficiency. Currently, global technology giants and large model companies are deploying AI agents.
Among them, OpenAI has Operator, Anthropic launches Artifacts and Computer Use, domestic Byte launches Button, Ali open source AgentScope, and Intelligent Spectrum’s AutoGLM, GLM-PC and other AI agent products.
Google CEO Sundar Pichai said that in the next 2-4 years, Agent Workflows will see significant progress. With this technology, AI can help users perform and ultimately complete the task after users describe the task in natural language. Google is advancing two related projects, one of which is that the browser extension Mariner AI can replace users by browsing the web, performing complex tasks, and providing complete results. He saidThe completion rate of this technology was only 50% a year ago, but now it has increased to 85%.
Intelligent AI CEO Zhang Peng recently told GuShiio.comAGI that in the past, technology may be regarded as a single direction. For example, the improvement of text technology can bring huge value. However, we now realize that to transform technology into productivity tools or two-way interactive robots, it cannot have obvious shortcomings, because humans are multimodal and very complex in handling problems and tasks, so the requirements for technical capabilities are comprehensive. In general, this type of AI agent technology is still in its early stages of development, but this technology is expected to quickly reach the maturity level of ordinary users.
“I think achieving this goal is completely feasible before the end of 2025. As for success rate, this is a relatively broad concept because it depends on the complexity of the specific task. rdquo; Zhang Peng said.
Regarding the Manus craze, Huatai Securities Research News pointed out thatIn terms of generalization capabilities to solve problems, the current major Agent products are still far from general Agents, including task planning and tool use. In the future, with the improvement of the capabilities of the underlying models, AI Agents are expected to achieve capabilities improvement in two aspects: First, in terms of task planning, the planning capabilities based on rules and parameters are gradually evolved to practice-based reflection and iteration; The second is to evolve from specific tools based on human configuration to the selection and planning of multiple tools, and even further create tools suitable for large models.
Industrial Securities emphasized that in the first year of the outbreak of AI Agents, companies are expected to boost performance by accumulating AI capabilities and increasing AI Agents ’empowerment of products and businesses. It is recommended to focus on tracking the progress of the AI Agent industry and seizing relevant investment opportunities.
