Chinese AI agent Manus is the newest breakthrough in autonomous AI technology
Manus groundbreaking achievement in creating general AI agent that is self-directing assistant that turns your thoughts into actions
Studying Manus reminds me of the early days of BabyAI from Yohei. I remember configuring a fork of BabyAGI an watching it in action—I was both shocked and excited. Now Manus evokes that same response. They’ve brilliantly leveraged freely available technology to create something truly stunning.
While AI agents like Manus are still in their infancy, this release has pushed the field forward significantly. Though this is clearly Manus’s most basic version—with more robust iterations to come—it breaks new ground. Current AI agents face a fundamental limitation: they require constant human supervision, functioning only under direct guidance. Like sophisticated puppets, they need human hands pulling their strings to animate their intelligence. Manus AI shatters this paradigm.
Developed by Chinese startup Butterfly Effect with Tencent’s financial backing, Manus positions itself as the world’s first fully autonomous AI agent. Its founder Xiao Hong, along with co-founder and chief scientist Yichao “Peak” Ji, has an impressive track record – after graduating from Huazhong University of Science and Technology, Hong founded Nightingale Technology in 2015, launched Butterfly Effect in 2022 with the successful Monica AI browser extension reaching 10 million users, and finally unveiled Manus AI in March 2025.
The distinction might seem subtle, but it’s revolutionary—imagine an AI that doesn’t just respond, but initiates, persists, and completes complex tasks without your continuous guidance.
Beyond conversation to continuous action
What separates Manus from chatbots like ChatGPT isn’t merely incremental improvement – it’s a fundamental shift in the relationship between humans and AI.
While ChatGPT excels at providing information or following step-by-step instructions, Manus operates more like a digital employee. It can break down complex projects, develop its own workflow, and execute over extended periods – even when you’re offline.
“The most impressive AI tool they’ve used,” is how experts from Hugging Face and top venture capitalists describe it. This isn’t hyperbole when you consider Manus can reportedly complete weeks of professional work in just hours.
The architecture of autonomy
Manus achieves this breakthrough through a multi-agent architecture that combines multiple AI models working in concert. This isn’t just one mind – it’s many specializations collaborating under a unified goal.
What’s fascinating is how Manus implements this multi-agent approach. When interacting with the system, users communicate exclusively with the “executor agent” – a specialized interface that doesn’t contain the details of the knowledge base, planner, or other specialized agents. This architecture brilliantly controls context length while enabling complex behaviors.
“Multi-agent implementation is one of Manus’s key features,” explains Peak from Manus AI. “Actually, it’s not that complicated. This really helps to control context length, and that’s why prompts obtained through jailbreaking are mostly hallucinations.”

Inside the sandbox
Each Manus session operates in its own sandbox environment, completely isolated from other sessions. Users can enter this sandbox directly through Manus’s interface, providing unprecedented visibility into how the AI operates.
The sandbox code primarily receives commands from agents and is only lightly obfuscated. While revolutionary in performance, Manus’s action space design follows established academic approaches rather than reinventing the wheel. Due to the RAG (Retrieval-Augmented Generation) mechanism, tool descriptions vary across different tasks.
The system integrates seamlessly with web browsers, code editors, and databases, giving it practical tools to accomplish real-world tasks. It handles text, images, and code with equal dexterity, making it versatile across domains.
Open source foundations
Perhaps most refreshing is Butterfly Effect’s commitment to the open source community. “We use many different open-source technologies, which is why I specifically mentioned in the launch video that Manus wouldn’t exist without the open-source community,” noted a company representative. This includes the well-known @browser_use open-source code among others.
The team has promised a series of acknowledgments and collaborations, with plans to open-source several components in the near future. “There’s no need to rush – our team has always had an open-source tradition, and I personally have been sharing my post-trained models on HuggingFace. We’ll be open-sourcing quite a few good things in the near future.”
The double-edged sword of true autonomy
This unprecedented level of independence raises fascinating questions. An AI system operating continuously without fatigue introduces ethical considerations that we’re only beginning to grasp.
The potential benefits are striking—picture complex data analysis running through the night, automated research gathering information as you sleep, or software writing itself to your specifications without needing constant oversight.
Yet significant challenges persist. The system’s current technical limitations lead to frequent crashes and timeouts during extended operations. Access also remains tightly controlled—fewer than 1% of waitlisted users have received invite codes.
The global AI race accelerates
Manus represents more than technological progress—it signals a shift in the global AI landscape. Through strategic partnerships, including its recent collaboration with Alibaba’s Qwen team, Manus is swiftly expanding its capabilities and reach.
This development showcases China’s rising prominence in AI innovation. While Western companies have dominated large language models, Manus marks a potential breakthrough in agent technology, advancing from conversational AI to truly autonomous systems.
Like all major technological advances, this raises important questions about ethics, security, and oversight. Yet the trajectory is undeniable—we’re moving from AI that simply responds to AI that takes initiative. Manus stands at the vanguard of this transformation.
Additional videos
Recent Blog Posts
Anthropic Shipped Claude Channels. Your AI Agent Can Now Text You Back.
Until very recently, every interaction with an AI agent had the same shape. You sit down. You open the tool. You give it a task. You wait. You check. You iterate. Every cycle requires your presence. Walk away and the session stalls, the output piles up unseen, or a permission prompt freezes everything until you come back. That constraint just changed. On March 20, 2026, Anthropic shipped a feature called Claude Code Channels. It lets Claude's agentic tool communicate with you through Telegram, Discord, and iMessage. You send a task from your phone. Claude does the work on your computer....
Apr 13, 2026What Did You Do Today?
There's a saying in Jackson Hole. You hear it at the coffee shop on the square, on the chairlift at the Village, in the bars after a day on the mountain. It goes like this: It's not what you do. It's what you did today. I've been thinking about that line all weekend. Because Sam Lessin dropped a piece arguing that AI isn't just a labor crisis — it's a meaning crisis. And Goldman Sachs just published 40 years of data proving that when technology displaces workers, the damage doesn't heal. It scars. Ten percent slower earnings growth for the...
Apr 3, 2026Claw-code Broke GitHub’s Star Record in 24 Hours. Two Engineers Did It on an Airplane. Here’s What That Means for Your Business.
Here's the number: 100,000. That's how many GitHub stars a repository called claw-code collected in roughly 24 hours. Not a year. Not a month. One day. By the time a live stream was done discussing it, the counter was climbing by a thousand stars every ten minutes. Nobody in the room could remember seeing anything grow that fast. Because nothing had. I watched it happen in real time. I'd met the two engineers behind it the weekend before at an AI hackathon in San Francisco. Within 72 hours of shaking hands, they'd built the fastest-growing repo in GitHub history —...