This week, Manus, a new AI agent from Chinese startup Monica.im, “went viral” capturing the attention of numerous AI commentators and polarising camps into the breathless and the sceptical. Marketed cleverly through scarce invite codes (ExoBrain is still waiting, somewhere in a supposed 2 million strong queue) and influencer-driven promotion, Manus rapidly dominated the headlines, labelled by some as another “DeepSeek moment” for China, or even a glimpse of AGI.
Yet, Manus isn’t ground-breaking in technological innovation terms in the same way R1 was. It’s essentially an interface built around Anthropic and Alibaba’s existing Claude and Qwen model, augmented with web browsing and command-line tools. Its strength lies not in pioneering new AI techniques, but in creatively enhancing the conversational interface users already know well.
Interestingly, Manus predominantly operates by executing code to complete varied tasks, an approach inspired by a 2024 research project known as CodeAct, where agents use executable Python code to unify their action space. CodeAct integrates a Python interpreter to dynamically adjust actions based on observations. While you’re waiting for your invite code, the Manus provide a series of recordings where you can see this code heavy process in action. The product is not without significant flaws. Users quickly exposed Manus’s basic security architecture, prompting the agent to share parts of its source code, revealing internal details and its reliance on Claude 3.7 Sonnet. Moreover, despite impressive demonstrations of tasks such as property research and financial analysis, many users found real-world performance inconsistent or exaggerated.
The Manus affair reveals important lessons about the types of innovation users actually want. While significant efforts go into refining chatbot windows or conversely experimenting with radical new interfaces such as Flora’s “intelligent canvas,” there’s a noticeable gap. Users often resist radically new paradigms due to the comfort of chat. Manus addresses this issue, materially extending the familiar chat model by embedding autonomous, agent-like features. The hype and criticism surrounding Manus highlight genuine user interest for conversational AI that moves beyond simple exchanges to achieve meaningful automation and practical productivity boosts. Manus’s approach, extending rather than reinventing conversational interfaces, offers a potential template for future UX innovation.
As AI policy commentator Dean Ball stated; “The Western AGI obsession makes us want to conceptualise [it] as one godlike model that can do everything, and we implicitly dismiss product engineering and practical applications. You see that reflected in public policy, which is obsessed with big models, giant datacentres, and similar infrastructure. Those are the only things we seem to take seriously and value.” Could it be that clever product design and not scale might unshackle AI?
Takeaways: Looking forward, Manus demonstrates the value of AI product engineering that explores the space between familiarity and novelty. As we reach the limits of refining traditional chatbots and encounter resistance adopting future paradigms, the “conversational-plus” space Manus occupies might become the centre of gravity for the next AI developments. Products like Google’s NotebookLM, Cursor for development, and now Manus suggest the next wave in AI won’t just rely on more powerful models, but rather smarter ways of using models to deliver agentic capabilities through interfaces we intuitively trust. Manus isn’t revolutionary, but its success signals clear demand for conversational AI that combines familiarity with genuine autonomy. Expect this product space—intuitive yet powerful conversational agents—to become a new frontier of practical innovation.