Chinese startup Monica, also known as Butterfly Effect AI, has developed a completely autonomous artificial intelligence agent named Manus AI. This engine was officially launched on March 6, 2025.
The Manus scripting language allows users to specify tasks in a more complex manner. The user can define subtasks (i.e., ask Manus to request assistance from a human). The user can set conditions (i.e., ask Manus what to do next if the human doesn’t obey).
The user can define in as much detail as they like what Manus should do if the human doesn’t obey and if it is in charge. Manus can then execute the user’s task with the same ease as a human. It can do it just as well with all its functions working from start to finish.
It utilises several AI models, including Anthropic’s Claude 3.5 Sonnet and a version fine-tuned by Alibaba, to tackle problems, generate plans, perform tasks (such as data scraping, coding, and web design), and provide answers.
Celebrated for its autonomy and GAIA standard-setting performance, Manus set off an international stir but also raised worries about privacy—for instance, who can get into it and who can’t—and, of course, concerns about what it can do (or will be able to do) and who can stop it.
Who is the founder behind Manus AI?
Manus AI was developed by the Chinese startup Monica, also known as Butterfly Effect AI, registered in Singapore.
It officially launched on March 6, 2025, and is celebrated as one of the world’s first fully autonomous AI agents capable of independent planning, reasoning, and executing real-world tasks.
A key figure associated with Manus is Xiao Hong, who is noted for previously founding Nightingale Technology and developing productivity tools integrated into platforms such as WeChat. He stepped down as a shareholder of Monica in December 2024.
The company’s public materials and coverage do not frequently cite a single “founder” by name beyond associating the broader team, with Monica itself being the recognised developer.
Today, Manus is recognised for its ambitious endeavour to bridge the “mind” and “hand”—its name derived from Latin, meaning “hands”—by autonomously executing complex, multi-step tasks in the cloud.
How does Manus AI Work?
Manus AI understands your request, breaks it into steps, and assigns tasks to specialised sub-agents. It works in the cloud, adapts to changes, fixes issues, and delivers precise, ready-to-use results—even when you’re offline.
1. Understands your request
You tell Manus what you want, like “plan a trip” or “analyse stocks”. It uses language models to grasp the goal and any key details.
2. Make a plan
Manus breaks your goal into smaller, ordered steps. It determines the order of tasks to complete first, next, and so on.
3. Uses many helpers (multi-agent setup)
Each step is handled by a specialised “sub-agent”: one for gathering info, another for browsing the web, one for writing code, and another for checking accuracy.
4. Works in the cloud—even when you’re offline
Once launched, Manus continues to run in the cloud. You can close your device, and it’ll still finish the job.
5. Fixes problems on its own
If something goes wrong—such as a flight being booked or data changing—Manus adjusts and finds a better solution without needing help.
6. Shows the process clearly
Unlike regular chat assistants, Manus reveals how it works behind the scenes—like “opening tabs”, “running code”, or “scraping websites”—so you can follow along.
7. Completes and delivers the final result
Once finished, Manus provides you with an organised output—such as a plan, report, website, or analysis—ready for use.
What are Its Key Features?
Manus AI operates as a fully autonomous general-purpose AI agent that interprets your prompt and handles complex, multi-step tasks end-to-end, without needing further human instruction.
Core Features:
Task decomposition + Multi-agent system
Manus breaks a big task into smaller subtasks and delegates each to specialised sub-agents. These can include planning, researching, coding, and execution parts that work in parallel or sequence.
Asynchronous cloud execution
Once it starts, Manus continues working even when you’re offline. It runs tasks in the cloud independently, eliminating the need for constant supervision.
Tool integration & real-world interaction
Manus can navigate websites, fill out forms, scrape data, interact with APIs, write code, and build outputs such as websites, reports, or presentations.
GAIA benchmark leader
Manus has achieved state-of-the-art results on the GAIA benchmark, scoring higher than OpenAI models across all difficulty levels.
“Wide Research” mult-agent expansion
On July 31, 2025, Manus launched Broad Research, enabling the orchestration of over 100 independent AI agents simultaneously for handling large-scale tasks, such as product analysis, content creation, and web scraping—all in parallel.
Conclusion
Launched in March 2025, Manus AI is a key turning point in artificial intelligence development. It is the first AI to act fully independently, turning intentions into actions without the need for constant human oversight.
In Manus, continuous, adaptive learning and efficient task execution are achievable because it is a multi-agent system.
By this, we mean that the system is composed of multiple intelligent agents that can interact with one another and with their environment. Yet, even a system as advanced as Manus has some challenges to overcome.
