For several months now, leading American developers of artificial intelligence systems have been talking about creating complex AI agents capable of performing certain tasks on behalf of the user. The little-known Chinese project Manus AI claims to have managed to get ahead of its overseas competitors, Bloomberg reports.
Image source: manus.im
Last week, Manus unveiled a universal AI agent, a service that reviews resumes, creates travel itineraries, and analyzes stock market performance at the user’s command. By some measures, the service performs better than OpenAI Deep Research, the Chinese company claims. Current AI agents still require significant user presence, while Manus’s system is “truly autonomous,” according to the project’s chief research officer, Yichao Ji. The company’s colorful video demonstration quickly went viral, generating a frenzy for invitations to test the service. Manus has been compared to DeepSeek, another Chinese startup that in January released an AI model capable of competing with leading American solutions. Manus has also raised questions about the United States’ leadership in AI, especially in a product category that American companies consider key.
The first reviews of the Manus AI agent were mixed. Professor Derya Unutmaz, a researcher in the field of immunotherapy in oncology, praised the system for the high quality of results, but noted that it processes tasks more slowly than OpenAI Deep Research. He was not the only one to complain about the speed – many noted that the service works slowly and often freezes before completing a task. This may be due to the limited computing resources of the company. In some cases, factual errors in the AI agent’s responses were also noted. “In reality, Manus is only a half-finished product,” said Yiran Chen, an American professor of electrical engineering and computer science who tried the service. The architecture of the system is still unknown, and it is not clear to what extent it uses models from other AI developers, because creating its own model would cost the company tens of millions of dollars.
The company behind the Manus project, Butterfly Effect, has raised more than $10 million in funding, several Chinese media outlets reported. Unlike DeepSeek, Manus has not published detailed documentation or explained how the agent was developed. Neither the service code nor the weights of the AI models have been published. The service user can choose between a standard mode and a “high effort” mode, in which requests are processed more slowly. Manus “creates a list of tasks, goes through the steps one by one, fixes problems that arise, asks questions if new commands are needed,” and gives an answer, OmniEdge founder Yong Qian commented on the mechanism of the system.
AI agents are currently being developed by OpenAI, Anthropic, and other industry companies — these systems go online, study online sources, and perform various multi-step tasks. Early adopters of Manus believe that the Chinese system is quite capable of competing with those currently on the market.