Photo credit: venturebeat.com
Stay updated with the latest insights and developments in the AI industry.
Recently, a relatively obscure Chinese startup has garnered international attention for unveiling a groundbreaking AI product.
This time, the spotlight is on Manus, a novel AI multipurpose agent designed not just to function as an AI model, but as an interface capable of managing multiple AI models. This agent can autonomously execute complex tasks such as generating comprehensive reports and managing several social media accounts on behalf of users.
If this sounds reminiscent of the advanced capabilities offered by AI giants like OpenAI and Google, particularly their Deep Research capabilities and other automation solutions, you’re not mistaken. Manus aims to consolidate these functionalities into a user-friendly platform.
What do business leaders and technology decision-makers need to know about Manus and its features? Let’s delve deeper.
What is Manus and who’s behind it?
Manus AI was officially unveiled on March 5, 2025, on the social networking platform X, where the company introduced it as “the first general AI agent” capable of executing advanced tasks autonomously rather than merely generating concepts.
As reported by the South China Morning Post (SCMP), Manus was created by a Chinese firm called Butterfly Effect, with offices in both Beijing and Wuhan. Although it has a small team of only a few dozen employees, it has quickly climbed the ranks in China’s burgeoning AI sector.
The founding group comprises seasoned entrepreneurs and product managers, spearheaded by Xiao Hong, a 33-year-old entrepreneur and 2015 alumnus of Wuhan’s Huazhong University of Science and Technology.
Xiao has a background in developing WeChat-based applications and was previously involved with Monica.ai, an AI assistant available as both a browser extension and mobile app.
According to the Manus website, the name “Manus” derives from Latin, meaning “hand,” symbolizing its function of assisting users in completing tasks.
How does Manus AI work?
Manus AI operates as a multi-agent system, integrating several AI models to independently manage tasks.
Unlike traditional AI chatbots that provide user assistance by offering information, Manus possesses capabilities to conduct research, analyze data, generate detailed reports, automate workflows, and even write and deploy software code.
As noted by Ji Yichao, co-founder and chief scientist of Manus AI, the platform is grounded in Anthropic’s Claude 3.5 Sonnet model—a model that has been available for nine months—alongside specialized versions of Alibaba’s Qwen models.
The team is currently working on upgrading Manus to utilize Anthropic’s latest and most advanced model, Claude 3.7, anticipated to significantly improve its reasoning and execution abilities.
Manus AI functions asynchronously, allowing users to delegate tasks and move on to other activities while the AI completes those tasks independently. Currently in private beta, access is granted through an invitation code system.
How does Manus AI stack up against U.S.-based competition?
A key reason for Manus AI’s rapid ascent is its impressive benchmark results, which reportedly place it ahead of alternatives like OpenAI’s o3-powered Deep Research agent, as reflected in data shared on its official site.
This performance has led enthusiasts and early users to believe that Manus may represent one of the most sophisticated autonomous AI agents accessible today.
Beyond benchmarks, Manus has demonstrated its effectiveness on freelance platforms such as Upwork and Fiverr, accomplishing intricate tasks typically requiring human intelligence.
AI influencers applaud Manus’s debut and remarkable performance
Discussion about Manus surged in media and AI circles last week as users on the platform X reported successfully employing it to manage up to 50 social media accounts simultaneously. This showcased its potential for creating robust engagement strategies for businesses.
While it remains unverified, the same technology could theoretically be leveraged for various marketing campaigns, including influence strategies and potentially even disinformation dissemination.
However, many AI experts in the West have expressed fascination with Manus, praising its initial performance and expressing awe at its capabilities during their exploratory use following acceptance of beta invites.
Rowan Cheung, founder of The Rundown AI newsletter, referred to Manus AI’s launch as a pivotal moment for AI agents, rebuffing the notion of it being “China’s second DeepSeek moment” in a LinkedIn post.
Cheung personally tested Manus, achieving results such as:
- Creating and deploying an accurate biography website with real-time data sourcing.
- Identifying prime rental locations in San Francisco based on crime metrics and the entrepreneurial landscape.
- Generating an extensive AI course consisting of eight comprehensive chapters.
After receiving 500 invitations from the Manus team, Cheung has started distributing them to his audience.
Bilawal Sidhu, a former Google employee and AI content creator, shared a comprehensive video review, dubbing Manus “the closest thing I have seen to an autonomous AI agent.”
He described the experience as akin to observing someone using a computer, guiding them at a high level while they execute tasks autonomously.
Sidhu evaluated Manus on several assignments, including:
- Conducting research on locations by scouring Google Maps and news articles to identify ideal spots based on accessibility and risk factors.
- Creating video applications that automate special effects and filters.
- Extracting market insights by aggregating data from platforms like Reddit and Twitter to generate in-depth reports.
He emphasized Manus’s ability to decompose complex workflows and implement them stepwise, positioning it as one of the leading autonomous AI agents available presently.
Another prominent AI commentator, Chubby (@kimmonismus on X), claimed Manus AI surpassed OpenAI’s Deep Research significantly, stating, “I do not get paid for promotion. I just tried preview access, and it outperformed OpenAI’s Deep Research by a lot! Mind-blowing.”
Andrew Wilkinson, co-founder of the startup exit provider Tiny, mentioned on X that he felt as if “he just time traveled six months into the future” after seeing Manus analyze 20 job applicants from a zip file, assessing each individually.
Big questions about Manus’s future remain
Amid the excitement, several apprehensions about Manus AI’s true capabilities have surfaced.
As reported by the South China Morning Post, the limited number of individuals able to test Manus AI is largely due to significant server shortages that arose following its rapid rise in popularity.
Zhang Tao, Manus AI’s product partner, acknowledged on social media that the current invite-only model is attributed to a genuine lack of server capacity, stating, “We underestimated the public’s response, and our server resources were only set for demonstration purposes.”
This scarcity has led to speculation that Manus AI might be leveraging artificial scarcity as a marketing strategy. Reports emerged of invitation codes selling for inflated prices on China’s second-hand marketplace, Xianyu, leaving many potential users frustrated.
Users such as @teortaxesTex on X, a vocal critic of AI products, have suggested, “it’s a product subtly tailored for influencers, which is likely why it gained so much attention. While it excels at generating popular content, it may not meet the needs of STEM professionals as effectively.” Even former VentureBeat journalist and current TechCrunch-er Kyle Wiggers remarked that “Manus probably isn’t China’s second ‘DeepSeek moment.’”
Is Manus AI the real deal?
Unlike DeepSeek R1, which created its own foundational model, Manus AI relies on existing large language models (LLMs) such as Claude and Qwen. This dependence has raised questions regarding the originality of Manus’s technology.
In an introductory video, Yichao Ji recognized this aspect, explaining, “Manus operates as a multi-agent system powered by several distinct models.”
Despite these concerns, Ji has promised a commitment to open-source development, with plans for some models to be released to the public within the year.
At its current stage in private beta, Manus AI is accessible only through invitation codes. Though the company aims to broaden access soon, server constraints may impede a complete public rollout. An open-source version is anticipated, allowing developers to implement components of its AI technology.
While some regard Manus AI as a significant advancement in autonomous AI agents, others remain doubtful about its genuine capabilities and scalability. As the enthusiasm surrounding Manus AI continues to swell, it raises the question: Will this be the next pivotal moment in AI, or simply another exaggerated claim in the evolving landscape of artificial intelligence?
Source
venturebeat.com