GuShiio.com ①Manus has become the new dark horse in the AI circle. This is a universal AI Agent that can “think” and “execute”;
② After Manus exploded, the evaluation of this product on the Internet began to polarization.
“Science and Technology Innovation Board Daily” March 7 (Reporter Zhang Yangyang)The tipping point of the AI circle is moving upward from the base model, this time to the application side-a product called Manus.
Manus, a product known as the first truly universal AI Agent, has swept the attention of the entire industry in the past two days.
The invitation code was hyped up to tens of thousands of yuan and one code was hard to find. After Manus became popular, the evaluation of this product on the Internet began to polarization. One party believes that this is a “DeepSeek moment for AI Agents” and believes that it demonstrates the rudiment of general artificial intelligence; the other party accuses Manus of over-marketing and the new product has just been unveiled, which has plunged the entire AI circle into a “scramble war”. Some people question its “marketing gimmicks are greater than technological breakthroughs”, and even bluntly say “shell innovation.”
丨 Universal AI Agent that can “think” and “execute”
Manus positions itself as a “general AI Agent”. It is not just an intelligent question and answer system, but has the ability to “think independently” and can independently complete multi-tasks and work collaboratively across platforms.
In a number of reviews, many users who received invitation codes tried tasks that were more complex and interesting than presentation videos. Some people used it to make a linktree-style personal homepage, or used Manus to solve practical problems for engineers, or asked Manus to make a minimalist chronicle of a country, or even asked Manus to make a video game… When the user completes the task in one sentence, it can produce a “finished product” with good quality.
“From the perspective of use cases, the functionality of being an AI Agent is very powerful and has a high degree of proficiency in dismantling tasks. The observation feeling of a CoA (agent chain) is very similar to that of looking at a CoT (thought chain), and it can ‘see’ AI in multiple scenarios. Evaluate and seek optimal solutions.” A technology blogger who received the invitation code test commented.
Another person who got the Captcha trial analyzed it from a technical perspective. The first core of Manus’s design was to equip the large model with a computer, allowing it to plan step by step for observation and action.
The second core is to assign it system permissions,”like opening up some company account permissions for new colleagues.” Manus has access to a large number of private APIs and can process a lot of structured authoritative data.
The third core is to train it, train it,”just like having a process of getting in with new colleagues.” Manus will also constantly learn your requirements based on your usage habits.
The source said that after a brief internal test, Manus still had some illusions and was still a little far from being truly universal.
丨 Shell or innovation?
Another controversy about Manus is the casing, whose modeling capabilities come from Anthropic’s Claude rather than the self-developed underlying large model.
On this point, the Manus team itself does not shy away from “shelving”. “‘Shell ‘itself is not terrible. The key is whether it can solve user problems.”Xiao Hong, founder and CEO of Manus, likened Big Model Companies to chip companies, while application companies to consumer electronics companies.
“Chip companies focus on producing higher-quality and cost-effective chips, while consumer electronics companies focus more on brands, channels and providing differentiated experiences for users. For consumer electronics companies, their core task is to create value for users, thereby achieving a closed loop of business.” Xiao Hong believes that, therefore,”sheathing” can be regarded as a way to integrate and optimize resources. As long as it can meet user needs and provide a good user experience, then “sheathing” is a successful product strategy.
An industry source said that from the perspective of technical implementation, the essence of Manus is “virtual machine + compute use + artifacts + multi-agent collaboration”. The core is to re-encapsulate existing Cursor products, lower the threshold for use, and allow non-technical users can also Make it easier to use AI to orchestrate complex tasks.
In other words, what Manus does is to add a “shell” to products like Cursor, and also build a batch of small models and Agents to achieve more automated reasoning, planning and execution capabilities. Although AI-assisted coding tools such as Cursor are powerful, they still have a high threshold for ordinary users to get started, and Manus is trying to further reduce technical barriers on this road.
However, Manus’s thinking is not entirely unique. For example, the OpenDevin team has tried a similar design. In essence, Claude 3 has strong processing capabilities for such tasks, and directly calling 20+ tools is a common operation. Therefore, from an “innovative” perspective, Manus did not bring a completely disruptive breakthrough, but rather further productization and optimization of existing ideas.
丨 AI Agent scale is approaching
In terms of comprehensive fermentation at multiple levels such as technology, user experience, and market sentiment, like the earlier DeepSeek, Manus is now rapidly becoming popular and becoming a dark horse in the current AI field.
From the perspective of business sustainability, its cost and commercialization have become another topic of discussion.
On the one hand, Manus currently maintains a low price advantage. Manus currently costs about $2 per task, which is far below the industry average. For example, ChatGPT Operator requires $200 a month for a Pro membership to use. And through open source reasoning modules and optimizing multi-agent collaboration efficiency, Manus’s costs are expected to be further reduced.
However, its commercialization still faces contradictions. According to a reporter from the Science and Technology Innovation Board Daily, the Manu team now uses AHPU (Agentic Hours Per User), the number of Agent runtimes used by a single user, as a measurement indicator.
Starting from Manus, AI Agents ‘consumption of tokens has increased exponentially compared to Chatbot, and their reasoning demand will increase 100- 1,000 times this year. Perhaps in the future, Manus can explore charging according to AHPU, but it needs to balance user experience and profit pressure. If other major manufacturers launch similar products at a lower cost, Manus’s first-mover advantage may be diluted. The team also admitted that there is only “3” A strategic window period “requires rapid iteration to build a moat.
Regardless of the praise and controversy, it is undeniable that Manus has slowly opened the door to AI applications. In the capital market, Manus’s popularity directly ignited market sentiment. On the day of the product’s release, more than 150 A-share AI agent concept stocks rose daily, and Cube Holdings and Coolte Intelligence rose by more than 20%.
Beijing Zhiyuan Research Institute predicts that 2025 will be the first year of large-scale implementation of AI Agents. Currently, AI competition is shifting from a “parametric arms race” to a “execution revolution.” As Xiao Hong said,”Manus is just a functional machine, and the real era of smart machines has not yet arrived.” In this transformation, the answer to who can become the “Apple in the AI era” may be hidden in every task execution on the user’s desktop.
