The "GPT Moment" of AI Agent, and the "universal hand" of users was born!
Reprinted from panewslab
03/06/2025·1MOriginal text: " The "GPT Moment" of AI Agent, Manus awakens the entire AI circle! 》
Author: shiyun Zhang Yongyi
Editor: Jingyu
2025 is the first year of AI Agent - this sentence was fulfilled in the early morning of March 6, Beijing time.
" After DeepSeek, another sleepless night in the tech circle."
Many users commented like this on social media.
Everyone stayed at night, just for the invitation code for the product - it is the world's first AI Agent product "Manus" developed by Monica.im.
According to the team, "Manus" is a truly independent AI agent that can solve various complex and changeable tasks. Unlike traditional AI assistants, Manus can not only provide suggestions or answers, but also deliver complete task results directly.
Manus\' introduction video is only 4 minutes long, but it is amazingly powerful | Image source: Monica.im
Just as the name "Manus" means, it symbolizes "hand" in Latin. In other words, knowledge must not only be in the mind, but also be able to be executed by hand. This is the essential advancement of Agent and AI Bot (chatbot) products.
Where is the Manus cattle? The most intuitive thing is to look at the use cases of official website display and user development demonstration. The Geek Park part is as follows:
- Travel Planning: Not only integrates travel information, but also creates customized travel manuals for users. For example, to plan your travel in Japan in April for users, provide personalized travel suggestions and detailed manuals.
- Stock Analysis: Conduct in-depth stock analysis and design visually attractive dashboards to show comprehensive stock insights. For example, conduct in-depth analysis of Tesla stock and create a visual dashboard.
- Education content creation: Create video demonstration materials for middle school teachers, explain complex concepts such as momentum theorems, and help teachers teach more effectively.
- Insurance Policy Comparison: Create a clear insurance policy comparison table to provide the best decision-making advice and help users choose the most suitable insurance product.
- Supplier Procurement: Do in-depth research throughout the network to find the supplier that best suits user needs and serve users as a truly fair agent.
- Financial Reporting Analysis: Capture market sentiment changes to specific companies (such as Amazon) through research and data analysis, providing market sentiment analysis for the past four quarters.
- Startup Company List Organize: Visit relevant websites to identify companies that meet the criteria and organize them into tables. For example, organize a list of all B2B companies in the YC W25 batch.
- Online store operation analysis: Analyze Amazon store sales data to provide actionable insights, detailed visualization and customization strategies to help improve sales performance.
- When the Agent finally outputs an extremely complete and professional result through a long list of thinking chains and tool calls, users began to sigh that "it can really help humans do things."
According to official website information, Manus has achieved new state-of-the-art (SOTA) performance on all three difficulty levels in GAIA benchmarks (evaluating the ability of universal AI assistants to solve real-world problems).
To sum it up - what Manus wants to do more is your literal "agent" in the digital world. And it did.
Just like you think, the Manus launched in the early morning suddenly woke up all the people in the AI circle!
01. Manus, your "digital agent"
First of all, Manus' experience is the biggest difference from previous LLM:
It emphasizes the ability to deliver the final result directly, rather than just giving a simple "answer".
Manus currently uses the Multiple Agent architecture, and runs in a separate virtual machine in a similar way to the Computer Use released by Anthropic. At the same time, various tools can be called in the virtual environment -writing and executing code, browsing web pages, operating applications, etc., to directly deliver the complete results.
In the official video, three work cases completed by Manus in actual use scenarios are introduced:
The first task is to filter resumes.
From 15 resumes, the appropriate candidates are recommended for the reinforcement learning algorithm engineer position and rank candidates based on their reinforcement learning expertise.
In this demonstration, you don't even need to decompress the compressed file or upload the resume files manually. Manus has already shown his side like a human "intern" by manually decompressing files, browsing each resume page by page, and recording important information in it.
 Manus, like an intern, automatically understands the hidden command \"uncompressing the packaging file thrown by the boss\" | Photo source: Geek Park
In the results given by Manus, not only are there automatically generated ranking suggestions, but it also divides candidates into different levels based on important dimensions such as work experience. After accepting that users prefer to present them in Excel tables, Manus can also automatically generate the corresponding tables by writing Python scripts on-site.
Manus can even record information such as "users prefer to accept results through tables" through memory ability in this practice. The next time a similar task result is processed, it will give priority to present it in the form of a table.
 Manus can remember users\' preferences in the content generation process | Image source: Geek Park
The second case, which is tailor-made for Chinese people, is to select real estate.
In the case, users want to buy real estate in New York. The input requirements are to have a safe community environment, low crime rates, and high-quality primary and secondary education resources - of course, the most important budget, enough to afford it with a fixed monthly income.
In this demand, Manus AI breaks down complex tasks into to-do lists, including researching safe communities, identifying quality schools, calculating budgets, searching for real estate, and more. And search the Internet to carefully read articles about the safest communities in New York and collect relevant information.
Second, Manus calculates an affordable property budget based on user income by writing a Python program. Based on the relevant housing price information on the real estate website, filter the property list based on the budget scope.
Manus can automatically search and filter out properties that do not meet user
requirements | Photo source: Geek Park
Finally, Manus will integrate all the collected information and write detailed reports, including community safety analysis, school quality assessment, budget analysis, recommended property lists and related resource links—like a professional real estate agent. Moreover, because Manus comes with the attribute of "completely based on user interests", its usage and experience are even better.
In the last case, Manus demonstrates analytical ability to stock prices.
The task given by the case is to analyze the correlation between stock prices of NVIDIA, Maywell Technology and TSMC in the past three years: it is well known that there is a close correlation between these three stocks, but for novice users, it is difficult to quickly sort out the causal relationship.
Manus's operation is very similar to a real stock broker. It first accesses information websites such as Yahoo Finance through APIs to obtain stock historical data, and at the same time cross-verify the accuracy of the data to avoid being misled by a single source of information, which has a significant impact on the final result.
In this case, Manus also used the ability to write Python code, perform data analysis and visualization, and also introduced professional financial-related tools for analysis. Finally, through data visualization charts and detailed comprehensive analysis reports, it feedbacks to users the causal relationships
- it is really like the daily work done by an "intern" in the financial field.
Not only that, Manus' official website also displays more than a dozen scenarios that can be used by Manus: directly use Manus to help you organize your itinerary, personalize the recommendation of travel routes, and let it learn to use various complex tools to complete daily work in a process-based manner.
In this process, what really makes Manus show that unlike the usual tools is its autonomous planning to ensure the ability to perform tasks.
The ability to learn independently also makes Manus’s work ability improvement logic more like a real human being—even at this stage, it may not be able to achieve expert level mastery in a specific field, but it can already see huge potential.
With the addition of independent learning ability, the generality of AI Agent has been greatly improved. In the actual test of user's Manus, you can even directly describe the relevant content in a video screen to it. Manus can ultimately directly follow the corresponding information, cross the limitations of platform content on search engines, and accurately find a link to a Douyin short video.
Since the current version of Manus is completely operated asynchronously on the cloud, Manus's capabilities are actually not limited by factors such as the end-side platform form or computing power you use - users can even temporarily shut down the computer after issuing instructions to Manus, and will automatically notify you of the results after Manus executes the activity result.
This operation logic is also very familiar - it's like after a person gets off work, he calls the intern on WeChat to "organize the documents and send them to me." However, now, this intern can really respond to you 7x24 hours a day, and there is no need to worry that he will "reorganize the workplace."
02. Multi-agent + self-check, run through AI Agent flow
From the above cases, it is not difficult to see that Manus's real killer weapon is not the concept of "AI Agent" that has appeared in Computer Use, but its ability to "simulate the way humans work."
Compared to "running computing", Manus's working logic is more like "thinking and executing commands". It does not do what humans really can’t do at the moment; that’s why some users who have experienced the current version of Manus describe it as “a trainee.”
On Manus' official website, there are many tasks that Manus can accomplish, and one of them shows how to use Manus in the B2B business. Quickly and accurately match your ordering needs with global suppliers.
In conventional products with similar needs, integrating global supply chain enterprise information within the platform to help users complete supplier/demand-side matching is a common logic in the industry. But in Manus' case, you can see a completely different way of implementing this.
Manus AI uses an architecture called "Multiple Agent" to run in a standalone virtual machine. Through the division of labor and cooperation mechanism of planning agents, executing agents, and verifying agents. to greatly improve the processing efficiency of complex tasks and shorten the response time through parallel calculations.
In this architecture, each agent may communicate with each other through an API or message queue based on a separate language model or reinforcement learning model. At the same time, each task is also run in the sandbox to avoid interfering with other tasks and support cloud expansion. Each independent model can mimic the process of human tasks, such as first thinking and planning, understanding complex instructions and breaking them into executable steps, and then calling appropriate tools.
In other words, through Manus' multi-agent architecture, it is more like multiple assistants who assist in retrieving resources, connecting, verifying whether the information is valid, etc., to help you complete the entire workflow - this is actually not only like you recruiting an "intern", but more like directly becoming a miniaturized "department director".
In the case of B2B business, Manus will automatically search in the vast ocean of the Internet through web crawlers and code writing and execution capabilities, and according to your own needs, it will match you with the most suitable source of goods from product quality, price, delivery capabilities, etc. Not only can you present the conclusions intuitively in front of you in a graphical way. More detailed operational suggestions can be given for these data.
 Manus fulfills the needs of B2B scenarios, which may be better than the built-in tools of a single platform | Image source: Geek Park
As for how and what technology the Monica team uses to achieve the video effect, according to the news, the team may announce it to everyone on March 6, Beijing time.
03. The ultimate of "stitching" is explosion
What kind of company is Monica.im behind Manus?
Monica is an All-in-One AI assistant, and its product form has gradually expanded from browser plug-ins to apps and web pages. The mainstream usage scenario is that when the user clicks its small icon in the browser, he can directly use the major mainstream models he accesses. Through an accurate understanding of the user needs of segmented scenes, Monica picked the "low-hunging fruit" of the big model.
Its founder Xiao Hong (nicknamed Xiaohong, English name Red) is a young serial entrepreneur born in 1992 and graduated from Huazhong University of Science and Technology. In 2015, he started his business after graduation, and his early entrepreneurship was not smooth (such as campus social networking and second-hand markets). In 2016, the operator of his WeChat official account provided editing and data analysis tools, gained millions of users and made profits. The final product was sold to a unicorn company in 2020.
After the 2022 big model wave, he officially founded Monica, focusing on overseas markets, and quickly completed a cold start through independent developer product ChatGPT for Google.
In 2024, Monica allows users to obtain the latest SOTA model as soon as the GPT-4o, Claude 3.5, and OpenAI o1 series are launched. With the new progress in the access model, the functions of professional search, DIY Bot, Artifacts writing applets, memory, etc. launched by Monica are also loved by users. Monica presents different interaction forms and functions in web pages with different functions such as YouTube, Twitter, Gmail, The Information, etc. to adapt to the user needs of specific scenarios, and has updated the personalized AI experience of hundreds of web pages.
In 2024, the number of Monica users doubled to 10 million. At the same time, it maintains considerable profits and ranks among similar overseas products.
Monica's strong performance proves one thing:
The shell is to the extreme, both TPF and PMF, and ultimately lead to user value.
 Monica Home | Photo source: Monica
Manus may continue this idea of the Monica team - Xiao Hong said in an interview with media person Zhang Xiaojun that the product cannot be just a form of chat robots, and the Agent will be a new form and need new products to undertake.
He took inspiration from AI programming products cursor and Devin. According to Geek Park, the former is mainly the copilot mode, and the latter is the autopilot mode, which is more in line with human needs. Agent should also be like Devin, facing the masses and truly led by ai. But the problem in the past is that the model is not smart enough.
However, based on the model, it is perhaps the advantage of the Monica team. Xiao Hong said that there are not many Agent product teams at present because it requires a lot of compound capabilities, such as the team having to do chatbot, AI programming, browser-related (because they all run on the browser), and having a good perception of the boundaries of the model - what level to develop today, what level will it develop next, etc.
"There are not so many companies with these abilities at the same time, and companies with these abilities may be doing a very specific business at hand, but we just happened to have time to do this together," he said.
Why did Monica do it? He concluded, "First, I think we are lucky. Second, to some extent, if everyone is doing reasoning today, may it be more time for startups? How far can the model expect the capability spillover to go?"
He believes that the Agent is still in its early stages. First, the Agent is still in the planning stage and has not yet reached the implementation of the physical world; second, the ability of the big model is still developing, and everything is still unpredictable.
" I definitely don't know that Agent can be found in this way. It's an unknown thing," he said.
What is interesting is that Monica, who "doesn't know how to make an Agent", has now made a product that makes the entire AI circle burst.
Manus may not be the final AI Agent, but it undoubtedly once again raised expectations for AI by orders of magnitude after DeepSeek's popularity.
*Head image source: Monica.im
chaincatcher