Image source: Generated by unbounded AI

AI also changes the sky again.

At the morning of March 6, a video of a product called Manus was flooded across the Internet. In the video, Peak, co-founder and chief scientist of Manus AI, said: Manus is the first universal AI agent.

A netizen sighed after watching the video, "Tonight's historical significance is no less than the night ChatGPT was released."

Since the morning of March 6, the official website of Manus has been in a state of collapse. There are people who are asking for invitation codes everywhere in the circle of friends and WeChat groups.

▲Screenshot of the collapsed Manus official website

In the afternoon of that day, the price of a Manus internal test code was called to 60,000 yuan on Xianyu.

▲The "outrageous" price of Manus invitation code on Xianyu

Many people sigh that Manus' voice is bigger than DeepSeek. So, as the "first general AI agent", what are the differences between Manus and the previous big model products?

Manus, which is not "nonsense", is in official terms, "This is not just a chatbot or workflow, it is a truly autonomous subject, bridging the gap between concepts and execution. When other artificial intelligence knowledge is producing ideas, Manus delivers results."

Speaking of the human thing is that other AIs are nonsense with you, Manus helps you work.

From an example from the official website, when the user asked Manus to help write a "7-day Japanese itinerary and proposal proposal." Manus will think like DeepSeek, and disassemble tasks and make a to-do list. Then it will start browsing web pages, collecting data, and organizing information like humans.

We can see that a small window will appear on the right side of the screen, which simulates the screen status and operation process seen by Manus.

▲Screenshot of Manus task processing

Even Manus will encounter verification code problems when accessing web pages and are displayed.

▲Manus encounters verification code when visiting the web page

After the relevant work is completed, Manus will output 6 files, including a 7-day itinerary, a romantic marriage proposal location, a Japanese travel research, an HTML file, and a to-do list (this is not something that the user needs to complete when going to Japan, but the work that Manus needs to complete in the process of making a plan, such as understanding the basic knowledge of traveling to Japan, studying cities and attractions, creating itineraries, etc.).

▲The work result document output by Manus

Take the most core "7-day itinerary" as an example, Manus lists in detail the daily itinerary and where discounts can be redeemed.coupons and other details.

▲Search of Manus' travel plan

For comparison, after we translated the same prompt words into Chinese, we fed them to DeepSeek, Tencent version of DeepSeek and Baidu version of DeepSeek respectively.

Among them, after deep thinking, the official version of DeepSeek outputs a 7-day itinerary, a specially recommended marriage proposal location, HTML travel manual template, budget allocation and other suggestions. The content is relatively simple, but the information is complete and available.

But DeepSeek cannot output files like Manus, and users need to paste the contents themselves.

▲DeepSeek official version task output results

As a comparison, Baidu version DeepSeek outputs HTML travel manual templates and budget allocations, and the content is relatively rough and not beautiful enough.

▲DeepSeek Baidu version task output results

After thinking about it, Tencent version DeepSeek only outputs HTML travel manual templates, and the layout is relatively messy.

We will find that Manus can perform multiple tasks at the same time and output documents that can be used directly.

▲DeepSeek Tencent version task output result

But if you want to say how much Manus is ahead of the big model commonly used now, it also needs to be compared.

For example, if compared with the Tencent version and Baidu version of DeepSeek, Manus's mission execution results are obviously far ahead. But if compared with DeepSeek's official official, Manus is actually better at the fact that the unified delivery of the documents are good. The same task can be completed well by the official version of DeepSeek.

Of course, the above is only one of the Manus cases. On Manus' official website, there are many similar cases, such as in-depth analysis of Tesla stocks, sorting out the list of YC companies, online store operation analysis, outputting a website SEO, etc.

From these cases, Manus has been able to help workers work in many work scenarios.

Manus's three "controversies"

But just as Manus was flooding the entire network, some controversies were always accompanied by Manus.

For example: cover?

After Manus came out, some netizens commented: Is it innovation to cover the shell to the extreme?

This comment obviously has some negative emotional color. In theory, Manus is really just combining big models and multiple small models on cursor products, that is, it really doesn't have much new things on the technical level.

But it is obviously not right to simply summarize the work they do as a "shell". In fact, many people underestimate the use of various large models andThe workload of integrating various agent + compute use capabilities.

▲Manus product release video

The second controversial point is the barrier.

Above mentioned, Manus does not have much innovation at the technical level. And DeepSeek and Claude can do things that Manus can do, and even OpenAI has developed similar products before.

There will be a question here: When Manus' products become popular, will they be quickly followed up by major manufacturers and seize the market? Will Manus's capabilities be internalized in the future by basic models such as OpenAI and Claude?

"I think Manus' barriers are limited."

Fang Yuan is a popular AI application entrepreneur who mentioned it to us after seeing Manus' products. "Compared with underlying R&D, product functions are easy to copy, such as byte changing the buttons." Fang Yuan said: "Because Manus is finally adjusting various applications to help you do things. Compared with startups, Byte and many large manufacturers have many application matrices, which have more advantages in this regard."

"I feel that after Manus, a large number of similar applications may come out soon," Fang Yuan said.

So, although Manus is another popular application after DeepSeek, unlike DeepSeek focusing on the research and development of basic large models, Manus relies on establishing applications based on the underlying large models to acquire users and realize a closed-loop business. Therefore, how to open up the living space between large manufacturers and model manufacturers may be a more realistic problem.

But it doesn't mean that Manus has no chance. "If it can handle this part of the traffic, Manus may become a traffic distribution place similar to the Agent app store, and do what major manufacturers such as Baidu and Microsoft have always wanted to do but have not done." Fang Yuan said.

▲OpenAI's third controversial point was how much value can the work completed by Manus create.

From the 50 cases on Manus' official website, it seems that it can already replace the work of "workers" in many scenarios.

For example, in the resume screening scenario, He Lin is a senior HR. After watching Manus's function demonstration, she told "Bordered Unknown" that this function is more effective during large-scale recruitment, especially the ability to automatically organize all resume information, which is very practical in their daily work and can save a lot of time.

But this does not mean that Manus can replace HR recruitment.

"Manus should be able to complete two tasks. One is preliminary resume screening, that is, to filter out some resumes through some fixed conditions, such as whether it is 985/211. Graduated, is the age under 35 years old, and has the work experience been over three years. ”

”After completing this part of the work, HR needs to manually screen the remaining resumes to judge those non-standard conditions, such as project experience, degree of matching with the position, etc. For an experienced HR, it will have more advantages in this regard. "He Lin said.

After completing this step, Agent will score all resumes and filter out the last few resumes and enter the interview stage to improve work efficiency.

▲Screenshot of Manus resume screening workflow

In addition to resume screening, another Manus application scenario known to "Bordered Unknown" is SEO optimization.

As a senior SEO, Wang Qiang first registered two domain names about Manus early in the morning of March 6. After reading Manus' demonstration case about SEO, Wang Qiang said that it can basically reach the level of junior and intermediate SEO.

But judging from the SEO optimization report on personal blogs output by Manus, what he accomplished is only a primary work. "The SEO optimization report it outputs is very detailed at many levels, but the value of this report only contains the diagnosis made by the website just established or the first time it is used to optimize SEO." Wang Qiang said: "Because its analysis framework is mechanized, which affects website SEO, in addition to fixed elements such as website structure, content quality, keyword layout, it is also related to the macro-public opinion environment on the Internet, the development background of the industry, etc.."

▲Manus outputs the website SEO optimization plan work screenshot

Therefore, for Manus, Wang Qiang's judgment is that it can currently complete the basic work, but if you want to rely on Manus to do SEO well and improve it, from the current demo, you cannot make a judgment.

Of course, the above judgment, whether it is HR's resume screening or website SEO optimization report, is only the judgment made by the two people based on their years of experience in their respective positions after watching the Manus demo, or after in-depth experience of the product, Manus can play more value in these tasks.

"Bordered UnKnown" learned through communication with multiple HRs that many large state-owned enterprises have begun to explore the combination of DeepSeek with internal recruitment systems, and explore the application and implementation of interview assistance, resume screening, performance management, etc.

The changes brought by Manus

Although from the current judgment, Manus is not like DeepSeek, which has brought so many disruptions to the industry, but the impact still exists.

"After seeing Manus,My first reaction is that I want to open up my software and make an interface that supports AI tool calls. "

Fang Yuan believes that many applications and websites in Manus may open interfaces for software calls like Manus. The earlier you do this, the easier it is for the product to be promoted. "In order to facilitate call, a data standard similar to the current communication protocol may be formed between products in the future. ”

In the future, every big model application may be a very vertical workflow. Software like Manus will become a center for calling various applications to complete tasks. The value of application entrepreneurs will evolve to do a good enough job in segmented scenarios so that those software that call workflows can be selected.

▲OpenAI founder Sam Altman introduced the Agent application platform

In addition, Fang Yuan also felt that Agent transformed the current corporate workflow. "The process of company development in the past is no longer suitable now. ”

In the past, a software was developed, first the product manager drew a sketch, and then handed it to the interaction designer. After the interaction designer sets the interaction, it was handed over to the visual designer. After this series of work was completed, the product manager then pulled the R&D review, discuss the development cycle, collaboration methods, etc.

Now this set of work processes is being completely subverted in the AI ​​era. With the support of AI, many links have been simplified and abandoned. Many distribution processes that originally required multiple roles to cooperate can work independently, such as front-end development, visual design, etc.

AI is indeed transforming our work and life more and more widely.

This phenomenon is gradually confirming that Sam, the founder of OpenAI The judgment Altman mentioned in his year-end summary at the end of 2024: “We believe in 2025 we will see the first AI agents joining the workforce and fundamentally change the company’s output. ”

Since 2025, various big model companies have also bet on Agents.

In early January, OpenAI announced the launch of a beta function called Tasks. Users only need to explain the needs and time to ChatGPT, and it can achieve automated processing. In early February, Microsoft CEO Nadella announced that GitHub Copilot will all-in agents and launch Microsoft's first independent SWE agent. In China, the latest press conference of Jieyuexing also announced the creation of an Agent cooperation ecosystem, which divides Agent into two categories: vertical Agent and smart terminal Agent, and cooperates with different ecological partners.

So, although Manus is very popular, it is definitely not an accident.

It can be almost foreseeable that after Manus, more and more Agent applications will continue to grow, bringing surprises after one after another.We can wait and see.