Chinese scientists have unveiled an artificial intelligence (AI) “agent” that makes its own decisions without requiring specific instructions from a human operator.
The AI agent, Manus, was developed by Chinese startup Butterfly Effect. Its representatives claim it is the world’s first general AI agent — meaning it demonstrates a level of autonomy that current AI models lack. The scientists who created Manus say it shows a potential glimpse of what artificial general intelligence (AGI) may one day be capable of.
This emerging type of AI responds to text prompts, similar to chatbots such as ChatGPT or DeepSeek. However, unlike chatbots, they work on different tasks, without the need for frequent, step-by-step instructions.
Manus isn’t available to the general public yet, but a limited distribution of invite codes has given a select few access, sparking a flurry of interest online. Some users appear to have created playable video games from simple prompts, while others have used Manus to design and launch websites.
It’s still early days for Manus, however, with some users reporting crashes and other issues such as a tendency to get stuck in an infinite feedback loop. The company is also aware of relatively high failure rates in Manus versus ChatGPT. Chief scientist Peak Ji acknowledged these as part of the teething issues in launching a new tool on the social media platform X.
Reporter Caiwei Chen, who had access to Manus for MIT Technology Review, wrote that the experience was “like collaborating with a highly intelligent and efficient intern.” However, Chen also noted that Manus sometimes lacked understanding of what it was supposed to do, made incorrect assumptions and cut corners. “Ultimately, it’s promising but not perfect,” Chen wrote.
Is this the dawn of AI agents?
Manus AI is the second Chinese AI tool to send shock waves through the tech industry this year. In January, DeepSeek changed the landscape of AI by achieving similar or better chatbot results than its American competitors for reportedly a fraction of the cost. Manus, on the other hand, could represent the first of an entirely new generation of AI.
Chatbots usually run on a single large language model (LLM), while Manus uses multiple LLMs and other independently operating software to work autonomously on a variety of tasks, MIT Technology Review reported. This is also known as a multi-agent architecture, in which multiple components communicate and collaborate to process tasks.
In a head-to-head between Manus and ChatGPT, the agent will often provide more detailed responses than the chatbot — a claim which the company has reinforced with testing data from the GAIA benchmark. However, Manus also takes a lot longer to provide those responses as it does deeper research, Tom’s Guide reported.
In other words, Manus will effectively go its own way and figure out what to do in response to a prompt, rather than relying on step-by-step instructions like a chatbot does. It’s designed to start tasks on its own and dynamically adjust its approach along the way, Forbes reported.
Give Manus a single prompt, and it will navigate the web, write code and analyze data for its reply, without requiring further intervention as a conventional LLM-powered chatbot would.
Some examples Manus AI has demonstrated on its website include planning a detailed holiday itinerary, analyzing the stock market and screening job resumes. The difference between this and a tool like ChatGPT is that Manus can break down and complete complex tasks without continuous input. The system works from the cloud, too, so users can close their computers at any time, and Manus will keep working in the background.
The arrival of a general AI agent brings a new wave of ethical questions and considerations surrounding the use of AI. Agents behave more like humans, but unlike humans, they can work fast and continuously — provided they don’t crash — and never get tired.
