ManusEver since it came out, it has been second only to DeepSeek on the traffic charts. Some say it rivals DeepSeek, some say it is the GPT moment for AI Agents, and at one point a single invitation code was being scalped for 50,000.
According to people familiar with the matter, Manus AI recently completed a round of financing led by Silicon Valley venture capital firm Benchmark, with the funding amount reaching $75 million (about 546 million RMB).
This financing increased Manus AI's valuation by about five times, raising it to nearly $500 million (about 3.644 billion RMB).
Although Manus is already famous, let me still introduce it to friends who have not heard of it: unlike conversational AI such as ChatGPT, Manus's core capability isbreaking down complex tasks into executable small steps, and calling tools (such as a browser and code editor) to autonomously complete the entire process, ultimately delivering the results directly to the user (such as reports, PPTs, and data analysis charts).
Today I will test just how good this legendary traffic king of AI Agents really is. Everyone, wait and see!
10 hands-on test cases
01. Mind map
Prompt: Make a mind map on the theme of what things AI can now do instead of humans.
Result:
We can see that Manus divided our task into 7 steps, slowly calling different tools step by step to complete it.
I think the completed effect is very good. This mind map shows the full picture of AI replacing human work from four different dimensions, and it is really quite comprehensive.
02. Table
Prompt: Find the official data and various reviews of the 10 hottest AI products right now, and based on the situation of each AI product, generate various rankings and display them in table form.
Result:
When you enter the prompt, it is best to check this option, so that Manus will improve your prompt by giving you some options, making the final result closer to your expectations.
I asked him to display the results in a table, and he directly gave me results beyond expectations. Not only were there various rankings displayed in tables (although the tables were in Word, not Excel), there was also a research report that was very comprehensively written. After reading it, you will know how to choose the large model you need.
03.PPT
Prompt: Create a detailed and highly readable research report on China's internet AI industry self-media, displayed in PPT form.Result:
He put together a 19-page PPT for me, including market size and growth trend analysis, major platform analysis, content creator ecosystem analysis, business model and profit method analysis, AI technology application case introductions, and future development trends.
It is quite comprehensive, but there is a bit of a problem with the layout of this PPT. I do not know if it is a problem with my prompt.
04. Document
Prompt: I am an AI self-media blogger. I want to give a speech on the theme of the advantages and disadvantages of AI Agents. Help me organize a detailed verbatim script and display it in document form.
Result:
I think both the structure and content of this document are good. If everyone reads this article as an AI beginner, they can still learn a lot. Moreover, I think the several points he made about the future development of AI overlap greatly with what I have in mind. I even suspect he has been monitoring me.
05. WeChat official account article
Prompt: I need a WeChat official account article about the origin, development, and future prospects of AI Agents, with a length of about 1,500 words, and it needs a clickbait-style title.
Result:
To be honest, he's not very good at being a clickbait writer. This kind of title can no longer make any waves in today's ocean of self-media articles. We bloggers are the ones who should be writing clickbait.
The article he wrote is also a bit shallow. It still feels a bit too AI-generated. It's the kind of article that feels no different from not reading it at all, too official, like reading a document.
06. Webpage
Prompt: Make a webpage for beautifying QR codes. Features include:
- Generate a QR code with a specified color and style based on the given content or link.
- Users can upload a QR code to change its color and style.
Result:https://xaplhfxg.manus.space
I think the implementation of this webpage's content is actually quite good. Basically all the functions in the prompt were implemented. The only issue is that after entering content in the input box, the displayed color is white, so I thought I hadn't entered anything.
07. Travel planning
Prompt: I am in Beijing. I want to travel to Jiuzhaigou. The departure date is May 1, and the return date is May 7. The approximate budget is 10,000 yuan. Please provide a detailed itinerary and a simple HTML travel brochure, including a map, descriptions of attractions, essential Japanese phrases, and travel tips we can refer to throughout the trip.
Result:
Something happened here that I think was very good. When he was confirming the details of the task with me, I noticed that daily phrases had been typed as Japanese phrases, and I corrected it in time. This feature made me like it even more.
I also really like the final result. The details are maxed out, from the map to the descriptions of attractions, to the specific itinerary and travel tips. If MCP could be integrated to help me figure out exactly which train to take and which guesthouse to book, that would be even better.
08. Teaching demonstration
Prompt: I am a chemistry teacher and am preparing to teach the silver mirror reaction. Could you create a series of clear and accurate demonstration animations and organize them into a simple HTML presentation?
Result:
During the generation process, Manus suddenly told me that it could not generate complex animations and hoped to replace them with better videos it found. I guess it is because the animation involved in my question was too complex. I have seen that it can make animations for the law of conservation of momentum that other bloggers made.
However, I think its behavior of asking me whether to use an alternative solution when it cannot do it is very good, because many AIs just keep working away and do not care at all whether the final thing can actually be used.
09. Stock analysis
Prompt: I want a comprehensive analysis of Moutai stock, including:- Overview: company profile, key indicators, performance data, and investment recommendations.
- Financial data: revenue trends, profit margins, balance sheet, and cash flow.
- Analyze market sentiment: analyst ratings, sentiment indicators, and news impact.
- Technical analysis: price trends, technical indicators, and support/resistance levels.
- Asset comparison: market share and comparison of financial indicators with major competitors
- Investment value: intrinsic value, growth potential, and risk factors
- Investment thesis: SWOT analysis and recommendations for different types of investors.
Result:
The entire article comprehensively covered the content in my prompt. As a stock market novice, this article should be sufficient, right? But what I find funniest is that at the end he added a disclaimer: This report is for reference only and does not constitute any investment advice. Investors should make independent judgments based on their own circumstances and bear the corresponding investment risks.
No way, no way! Surely no one would actually change their investment direction and amount because of content generated by an ordinary AI, right?? Don't really treat AI as an industry insider. AI hasn't reached that level yet.
10. Popular Science Explanation
Prompt: Please explain the Battle of Shangganling to me in detail.
Core requirements of the explanation: describe the course of the battle in the form of a "battle report," and combine maps and terrain visualization, using specific maps to display and explain the process of the battle.
Result:
This explanatory article is really detailed, and it is also accompanied by operational process maps for each stage (though I can't understand them either), but the text content is very substantial. It includes the battle background, the organization of participating troops, the course of the battle, analysis of key battle points and terrain, casualty statistics, and the battle's impact and conclusions. I had not really learned much about this battle before, but after reading it I also roughly understood the process of this battle.
Summary
After testing these 10 cases, Manus is indeed quite impressive.
As the world's first general-purpose AI Agent, it has indeed achieved being smarter and more comprehensive!
It does not require complex, professional prompts, and it will also accurately understand our needs through "confirming problem details," autonomously completing the entire process from task decomposition to delivery.
Manus's breakthrough is not in multimodal capabilities, but in achieving a qualitative change at the cognitive level. AI has upgraded from a passively responsive tool to an actively executing intelligent agent.
- Multi-agent writing system: While other Agents are still imitating human operations, Manus has already built a unique "think-verify-execute" closed loop.
- Deep toolchain integration: The Manus tool library supports intelligent invocation of 200+ types of tools, giving it the ability to complete various tasks such as information gathering, coding, and data analysis.
- Dynamic evolution capability: User feedback can promptly optimize model strategies. Manus adopts a hierarchical memory network to achieve reuse of cross-project experience during task execution.
If Manus can continuously optimize its execution capabilities and expand irreplaceable application scenarios, it may become an important milestone in the commercialization of AI Agents. Gartner predicts that by 2028, 15% of daily work decisions will be completed by intelligent agents, and Manus's "digital intern" role is paving the way for this future.
Original article link:In-depth Manus experience: what can a general-purpose Agent do? 10 cases tell you