Original article: " AI Agent's "GPT Moment", Manus woke up the entire AI circle! "
Author: Shiyun Zhang Yongyi
Editor: Jingyu
2025 is the first year of AI Agent - this statement came true in the early morning of March 6th, Beijing time.
"After DeepSeek, another sleepless night for the technology community."
Many users made such comments on social media.
Everyone stayed up all night just for an invitation code to use the product - it was "Manus", the world's first AI Agent product developed by Monica.im.
According to the team, “Manus” is a truly autonomous AI agent that can solve a variety of complex and changing tasks. Unlike traditional AI assistants, Manus can not only provide suggestions or answers, but also directly deliver complete task results.
Manus’ introduction video is only 4 minutes long, but it’s incredibly powerful. Image source: Monica.im
As the name "Manus" implies, it symbolizes "hand" in Latin. In other words, knowledge should not only be in the brain, but also be executed by hand. This is the essential advancement of Agent and AI Bot (chatbot) products.
What is Manus good for? The most intuitive way is to look at the official website and the use cases spontaneously demonstrated by users. Geek Park has compiled some of them as follows:
- Travel planning: Not only does it integrate travel information, but it also creates customized travel guides for users. For example, it helps users plan a trip to Japan in April and provides personalized travel suggestions and detailed guides.
- Stock Analysis: Conduct in-depth stock analysis and design visually appealing dashboards to present comprehensive stock insights. For example, conduct an in-depth analysis of Tesla stock and create a visual dashboard.
- Educational content creation: Creating video presentations for secondary school teachers that explain complex concepts like the momentum theorem, helping teachers teach more effectively.
- Insurance policy comparison: Create clear insurance policy comparison tables and provide the best decision suggestions to help users choose the most suitable insurance products.
- Supplier Sourcing: Conduct in-depth research across the entire network to find the suppliers that best suit your needs and serve you as a truly impartial agent.
- Financial Report Analysis: Capture changes in market sentiment toward specific companies (such as Amazon) through research and data analysis, providing market sentiment analysis for the past four quarters.
- Startup company list compilation: Visit relevant websites to identify eligible companies and organize them into a table. For example, compile a list of all B2B companies in the YC W25 batch.
- Online Store Operations Analytics: Analyze Amazon store sales data to provide actionable insights, detailed visualizations, and customized strategies to help improve sales performance.
- When the Agent goes through a long chain of thinking and tool calls and finally outputs an extremely complete and professional result, users begin to exclaim that "it can really help humans do things."
According to the official website, Manus achieved new state-of-the-art (SOTA) performance in all three difficulty levels in the GAIA benchmark, which evaluates the ability of general AI assistants to solve real-world problems.
To sum it up in one sentence - Manus wants to be your "agent" in the digital world in a literal sense. And it has done it.
Just as you might imagine, the launch of Manus in the early morning woke up everyone in the AI circle!
01. Manus, your digital agent
First of all, the biggest difference between Manus and the previous LLM in terms of experience:
It emphasizes the ability to directly deliver end results rather than just giving a simple "answer".
Manus currently uses a Multiple Agent architecture, and its operation mode is similar to Computer Use released by Anthropic, and it runs completely in an independent virtual machine. At the same time, various tools can be called in the virtual environment - writing and executing code, browsing the web, operating applications, etc., and the complete results can be delivered directly.
In the official video, three cases of work completed by Manus in actual usage scenarios are introduced:
The first task is to screen resumes.
Recommend suitable candidates for the reinforcement learning algorithm engineer position from 15 resumes and rank the candidates based on their reinforcement learning expertise.
In this demo, you don’t even need to unzip the compressed file and manually upload the resume files one by one. Manus has already shown its side like a human “intern”, manually unzipping the file and browsing each resume page by page, while recording the important information in it.
Manus, like an intern, automatically understood the hidden instruction "unzip the package file thrown by the boss" | Image source: Geek Park
In the results given by Manus, there are not only automatically generated ranking suggestions, but it also divides candidates into different levels based on important dimensions such as work experience. After receiving the user's preference for presentation in Excel form, Manus can also automatically generate the corresponding table by writing Python scripts on the spot.
Manus can even use his memory to record information such as "users prefer to receive results in a table format" during this practice. The next time he deals with similar task results, he will give priority to presenting them in a table format.
Manus can remember user preferences in the content generation process | Image source: Geek Park
The second case, which is more tailored for Chinese people, is real estate selection.
In this case, the user wants to buy a property in New York, and the requirements he enters are that he wants a safe community environment, a low crime rate, and high-quality elementary and secondary education resources - and of course the most important budget, which is enough to afford it with a fixed monthly income.
In this demand, Manus AI breaks down complex tasks into a to-do list, including researching safe communities, identifying quality schools, calculating budgets, searching for properties, etc. It also collects relevant information by searching the web and carefully reading articles about the safest communities in New York.
Secondly, Manus wrote a Python program to calculate the affordable property budget based on the user's income. Combined with the relevant house price information on real estate websites, the property list was filtered according to the budget range.
Manus can automatically search and filter out properties that do not meet user requirements | Image source: Geek Park
Finally, Manus will integrate all the collected information and write a detailed report, including community safety analysis, school quality assessment, budget analysis, recommended property list and related resource links - just like a professional real estate agent. And because Manus comes with the attribute of "completely based on user interests", its use experience is even better.
In the final case, Manus demonstrated his ability to analyze stock prices.
The task given in the case is to analyze the correlation between the stock prices of NVIDIA, Marvell Technology, and TSMC over 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 cause and effect relationship.
Manus's operation is very similar to that of a real stockbroker. It first accesses information websites such as Yahoo Finance through API to obtain historical stock data. It also cross-verifies the accuracy of the data to avoid being misled by a single source of information, which would have a significant impact on the final results.
In this case, Manus also used the ability to write Python code, perform data analysis and visualization, and also introduced professional financial tools for analysis. Ultimately, through data visualization charts and detailed comprehensive analysis reports, he provided users with feedback on the causal relationship - really like the daily work of an "intern" in the financial field.
Not only that, the Manus official website also displays more than a dozen scenarios where Manus can be used: you can directly use Manus to help you organize your itinerary, recommend personalized travel routes, and let it learn to use various complex tools to complete your daily work in a streamlined manner.
In this process, what really makes Manus different from previous tools is its ability to plan autonomously to ensure the execution of tasks.
The ability of autonomous learning also makes Manus's work ability improvement logic more like that of a real human - even though it may not be able to achieve expert-level proficiency in a specific field at this stage, it can already see great potential.
With the addition of autonomous learning capabilities, the versatility of AI Agent has been greatly improved. In actual user tests on Manus, you can even directly describe the relevant content in a video screen to it. Manus can eventually directly find the link to a certain Douyin short video based on the corresponding information, transcending the restrictions of the platform content on search engines.
Since the current version of Manus runs completely asynchronously in the cloud, Manus's capabilities are not actually limited by factors such as the terminal platform form or computing power you use - users can even temporarily turn off the computer after giving instructions to Manus, and when Manus completes the activity results, it will automatically notify you of the results.
This operation logic is also very familiar - just like a person who calls the intern on WeChat after get off work to "send me the documents after they are sorted out". However, now, this intern can really respond to you 24/7, and you don't have to worry about him "rectifying the workplace".
02. Multiple agents + self-checking to run AI Agent flow
From the above cases, it is not difficult to see that the real killer feature of Manus is not the concept of "AI Agent" that has appeared in Computer Use, but its ability to "simulate the way humans work."
Rather than "running calculations", Manus's working logic is more like "thinking and executing commands". It does not do anything that humans cannot currently do; this is why some users who have experienced the current version of Manus describe it as "an intern".
The Manus official website displays many tasks that Manus can accomplish, including a case study that shows how to use Manus in B2B business. Quickly and accurately match your order needs with global suppliers.
In conventional products with similar needs, it is common logic in the industry to integrate global supply chain enterprise information within the platform to help users complete supplier/demand matching. But in the case of Manus, you can see a completely different way of implementation.
Manus AI uses a set of architectures called "Multiple Agents" and runs in independent virtual machines. Through the division of labor and cooperation mechanism of planning agents, execution agents, and verification agents, it can greatly improve the processing efficiency of complex tasks and shorten the response time through parallel computing.
In this architecture, each agent may be based on an independent language model or reinforcement learning model, and communicate with each other through APIs or message queues. At the same time, each task is run in a sandbox to avoid interfering with other tasks, while supporting cloud expansion. Each independent model can imitate the process of human task processing, such as thinking and planning first, understanding complex instructions and breaking them down into executable steps, and then calling the appropriate tools.
In other words, through Manus's multi-agent architecture, it is more like multiple assistants, who assist you in completing tasks such as retrieving resources, connecting, and verifying whether the information is valid, to help you complete the entire workflow - this is actually not only like you have hired an "intern", but more like directly becoming a miniature version of a "department head".
In the case of B2B business, Manus uses web crawlers and code writing and execution capabilities to automatically search the vast ocean of the Internet and match you with the most suitable source of goods based on your needs, including product quality, price, delivery capacity, etc. It can not only present the conclusions in a graphical form, but also provide more detailed operational suggestions for these data.
Manus meets the needs of B2B scenarios and may be better than the built-in tools of a single platform | Image source: Geek Park
As for how and what technology Monica's team used to achieve the video effects, according to reports, the team may reveal it to everyone on March 6, Beijing time.
03. The ultimate of "stitching" is explosion
What kind of company is Monica.im, the company behind Manus?
Monica is an all-in-one AI assistant. Its product form has gradually expanded from browser plug-ins to apps and web pages. The mainstream usage scenario is that when users click on its small icon in the browser, they can directly use the major mainstream models it accesses. By accurately understanding the needs of users in segmented scenarios, Monica has picked the "low-hanging fruit" of large models.
Its founder Xiao Hong (nickname 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 own business after graduation. His early entrepreneurial ventures were not smooth (such as campus social networking and second-hand markets). In 2016, he started a WeChat public account operator to provide editing and data analysis tools, which gained millions of users and achieved profitability. The final product was sold to a unicorn company in 2020.
After the big model wave in 2022, he officially founded Monica, focusing on overseas markets, and quickly completed the cold start of the product through the independent developer product ChatGPT for Google.
In 2024, Monica will allow users to access the latest SOTA models as soon as GPT-4o, Claude 3.5, and OpenAI o1 series are launched. With the new progress of access models, Monica's professional search, DIY Bot, Artifacts app writing, memory and other functions are also popular with users. Monica presents different interactive forms and functions in web pages with different functions such as YouTube, Twitter, Gmail, and The Information to adapt to user needs in specific scenarios, and updates the personalized AI experience of hundreds of web pages.
In 2024, the number of Monica users will double to 10 million. At the same time, it maintains considerable profitability and ranks first among similar products overseas.
Monica's strong performance proves one thing:
When the shell is taken to the extreme, it is both TPF and PMF, and ultimately leads to user value.
Monica Home | Image source: Monica
Manus may have continued the idea of Monica's team. When Xiao Hong was interviewed by media person Zhang Xiaojun, he said that the product cannot only be a chatbot. Agent will be a new form that requires new products to take over.
He got inspiration from the AI programming products cursor and Devin. According to Geek Park, the former is mainly a copilot mode, while the latter is an autopilot mode, and the latter is more in line with human needs. Agent should also be like Devin, facing the general public and truly led by AI. But the problem in the past was that the model was not smart enough.
However, the ability to package scenarios based on the model may be the advantage of Monica's team. Xiao Hong said that there are not many Agent product teams at present, because it requires a lot of complex capabilities, such as the team must have experience in chatbot, AI programming, browser-related (because everything runs on the browser), and have a good understanding of the boundaries of the model - what level it has developed to today, what level it will develop to next, etc.
"There are not many companies that have all these capabilities at the same time, and those that do may be working on a very specific business, but we happen to have classmates who have the time to work together to get this done," he said.
Why did Monica come up with it? He concluded, "First, I think we are lucky. Second, to some extent, if everyone is doing reasoning today, maybe there will be more time for startups? How far can the model predict the spillover of capabilities?"
He believes that Agent is still in its early stages. First, Agent is still in the planning stage and has not yet been implemented in the physical world; second, the capabilities of large models are still developing and everything is unpredictable.
"I definitely don't know that Agent can be taken out in this way. It's an unknown thing," he said.
What is interesting is that Monica, who “didn’t know how to be an agent”, has now created a product that has shocked the entire AI community.
Manus may not necessarily be the final AI agent, but it has undoubtedly raised people's expectations for AI by an order of magnitude after the popularity of DeepSeek.
*Header image source: Monica.im
