- Chinese researchers say they’ve built the world’s first fully autonomous AI agent called Manus AI.
- Its creators say Manus can perform tasks, like building a website from scratch, with minimal oversight.
- Some are calling it China’s next DeepSeek moment, but others are calling it overhyped.
Chinese AI researchers say they’ve created the world’s first fully autonomous AI agent — but others aren’t convinced that its capabilities make it another “DeepSeek moment.”
The agent, called Manus, launched last week and quickly became a hot topic in AI circles.
Its viral launch demo sparked a wave of online discussion, with some praising its capabilities and others pointing out its flaws, along with raising concerns about privacy.
Here’s what you need to know.
‘Completely autonomous agent’
Manus was built by the Chinese startup Monica, according to multiple reports, which appears to be a subsidiary of Butterfly Effect. Manus’ privacy policy states that Butterfly Effect is a Singapore-registered entity.
Unlike most traditional chatbots, which can require multiple human commands to complete tasks, Monica claims Manus can handle complex tasks independently from start to finish with just an initial user prompt.
Its launch video appears to show Manus AI sorting résumés, ranking candidates, and formatting data into spreadsheets in seconds.
Manus AI / Youtube
“This isn’t just another chatbot or workflow tool,” Yichao ‘Peak’ Ji, the cofounder of Manus, said in the Wednesday video. “It’s a completely autonomous agent that bridges the gap between conception and execution.”
The company says Manus can analyze stock trends, scrape data from the web, and even generate interactive websites from scratch. It runs in the cloud, meaning it keeps working even when users log out.
According to Manus’ website, it outperforms OpenAI’s Deep Research model on the GAIA benchmark, a tool for comparing models.
For now, access to Manus is limited, requiring an invite to test.
How was Manus created?
Following its launch, some early users suggested Manus wasn’t built from scratch and that it’s more of an AI wrapper running on other models like Anthropic’s Claude 3.5 Sonnet.
Pierre-Carl Langlais, the cofounder of AI startup Pleias and early Manus user, said in a Sunday LinkedIn post that Manus used token conventions that are “completely specific to Anthropic and not used by any other” large language models, or LLMs.
While the launch video hinted at multiple distinct models powering Manus, Ji later clarified that it currently uses Claude 3.5 Sonnet v1 and fine-tuned Alibaba’s Qwen models.
“We use Claude and different Qwen-finetunes,” Ji posted on X Monday. “Back when we started building Manus, we only got Claude 3.5 Sonnet v1 (not long-CoT, aka reasoning tokens), so we need a lot of auxiliary models. Now Claude 3.7 looks really promising, we are testing internally, will post updates!”
Manus draws DeepSeek comparison
Some have drawn comparisons between Manus and DeepSeek, the Chinese startup that in January unveiled an AI model with capabilities rivalling Western competitors despite appearing to be built at a fraction of the cost.
The launch spooked investors, triggering a $600 billion sell-off in Nvidia stock.
It also showed that China was capable of building leading AI models despite US export controls limiting its access to advanced chips, causing some to call it China’s “Sputnik moment.”
Now, Manus is stepping into the spotlight with claims of full autonomy — adding to the narrative that China is closing the gap with the US in the AI race.
Dean Ball, an AI policy researcher, said in a Sunday X post that it’s “wrong to call Manus a ‘DeepSeek moment'” — because it goes even further.
“DeepSeek was about replicating capabilities already achieved by American firms,” Ball said. “Manus is actually advancing the frontier. The most sophisticated computer using AI now comes from a Chinese startup, full stop.”
Researchers are divided
AI researchers and founders are split about Manus’ capabilities and what it means for the industry.
Some believe it’s a real breakthrough.
Hugging Face’s head of product, Victor Mustar, called Manus “the most impressive AI tool I’ve ever tried,” adding that its agentic capabilities “redefine what’s possible.” In a post on X, he suggested the tool could eliminate the need for coding as we know it. “This could kill vibe coding… It’s more like ideas vibing or name it however you want.”
Not everyone is convinced. TechCrunch’s Kyle Wiggers and Pleias cofounder Alexander Doria said they found Manus prone to factual errors, execution failures, and endless loops during tests.
Langlais had mixed reactions after testing it. “Overall, the product itself is not bad, especially on the UI side, giving the possibility to go back and forth checking whatever Manus is doing,” he wrote on LinkedIn.
But he called out the “deceptive communication” around Manus and what he described as “hunger marketing” tactics — manufacturing hype by limiting access to select influencers.
“What the AI market really needs now is better standards of openness and transparency. At every level: model, data, business,” Langlais said.
Manus has privacy concerns
One thing that Manus does share in common with DeepSeek is concerns about data privacy.
AI researcher Luiza Jarovsky raised concerns about where Manus stores its data and whether Chinese authorities have access — something that’s being fuelled by Manus’ opaque ownership.
“Where are the servers located? Is user data transferred to China?” Jarovsky asked in a Substack post.
Mel Morris, CEO of Corpora.ai, sees promise in Manus but isn’t convinced it’s a major breakthrough — and points to data privacy concerns due to its ties to China.
“These factors will only intensify the ongoing debate over AI security and governance.”
A spokesperson for Manus did not immediately respond to a Business Insider request for comment.
