The Emergence of Manus and its Capabilities
The rapidly evolving field of AI agents has recently seen a flurry of activity. A new startup from Shenzhen, China, called Manus, has introduced what it terms a “general purpose AI agent,” sparking considerable discussion and excitement within the AI community. Manus’s creation is designed to function as a versatile AI agent, capable of autonomous planning, execution, and delivery of comprehensive results. This agent interacts with websites in real-time, processes various data types, and utilizes a suite of tools to achieve its objectives.
Despite being in an invite-only phase, Manus quickly gained attention for its impressive capabilities. Deedy Das, a principal at Menlo Ventures, lauded Manus, stating, “Manus, the new AI product that everyone’s talking about, is worth the hype. This is the AI agent we were promised.” Das highlighted the agent’s ability to condense what would typically be two weeks of professional work into approximately one hour.
Andrew Wilkinson, co-founder of the technology holding company Tiny, expressed a similar sentiment, remarking, “I feel like I just time travelled six months into the future.” Wilkinson even shared that he tasked Manus with developing and substituting a software solution for which his company currently spends $6,000 annually.
Manus has showcased a wide array of functionalities, including:
- Detailed Itinerary Creation: Generating comprehensive travel plans.
- In-Depth Data Analysis: Performing thorough analysis of stocks and businesses.
- Research Report Generation: Producing reports on diverse topics.
- Game Design: Conceptualizing and designing games.
- Interactive Educational Courses: Developing engaging learning experiences.
Users have described Manus as a multifaceted tool, combining deep research capabilities, autonomous operation, computer use functionality, and a coding agent equipped with memory.
User Experience and Performance Benchmarking
Beyond its “mind-blowing” agentic capabilities, as some have put it, Manus has also been commended for its user experience (UX). Victor Mustar, head of product at Hugging Face, noted, “The UX is what so many others promised, but this time it just works.” The design of Manus also incorporates human oversight, requiring approvals and permissions for various actions.
Manus has also been put to the test on the GAIA benchmark, which evaluates general AI assistants on their ability to solve real-world problems. According to the reported results, Manus demonstrated superior performance compared to OpenAI’s Deep Research. The GAIA benchmark is a challenging test, and Manus’s performance suggests a significant level of competence in handling complex, real-world scenarios. This benchmark provides a quantitative measure of Manus’s capabilities, complementing the qualitative praise it has received from early users.
The ‘Wrapper’ Debate and the Value of Manus
A few days after the initial wave of excitement, some users on X (formerly Twitter) discovered that Manus was operating on top of Anthropic’s Claude Sonnet model, alongside other tools like Browser Use. This revelation led to some expressions of disappointment, with some critics suggesting that Manus lacked a unique “moat” or competitive advantage. The argument was that, since Manus relies on pre-existing models and tools, it doesn’t offer anything fundamentally new.
The reality is that Manus, to achieve its impressive capabilities, functions as a “wrapper” around some of the most advanced AI models available. This approach, however, has sometimes been met with an oddly negative perception on social media. Some users seem to view the “wrapper” approach as inherently less valuable than creating a completely new model from scratch. This perspective overlooks the significant engineering effort and design choices involved in effectively integrating and orchestrating existing AI components.
Ultimately, Manus has demonstrated success in creating a well-designed interface that effectively harnesses the agentic potential of a foundational AI model. It’s not simply about using Claude Sonnet; it’s about how Manus uses it, along with other tools, to achieve its impressive results. The “wrapper” criticism misses the point that innovation can come in many forms, including the clever combination and application of existing technologies.
Aidan McLaughlin, a professional at OpenAI, commented on X that the “wrapper” aspect was not a significant concern. He emphasized, “If it created value, it deserves my respect. Care about capabilities, not architecture.” This statement highlights a crucial point: the ultimate measure of any technology is its ability to deliver value to users. Whether that value is achieved through a novel architecture or a clever combination of existing components is secondary.
Furthermore, initial reviews of Manus highlight the untapped potential of current AI models, capabilities that even the labs developing them haven’t fully realized. Richardson Dackam, founder of GitGlance.co, stated, “Manus didn’t just slap an API on a model. They built an autonomous system that can execute deep research, deep thinking, and multi-step tasks in a way that no other AI have.” This underscores the point that Manus is not merely a superficial interface; it represents a significant advancement in how AI models are utilized to perform complex tasks.
The Question of Unreleased Capabilities
The success of Manus, built upon existing models, raises an intriguing question: if Manus is built upon existing models from the United States, why haven’t the creators of those models been able to deliver similar capabilities themselves? This question points to a potential gap between the theoretical capabilities of AI models and their practical application in user-facing products.
Dean W Ball, an AI researcher, suggested, “I assume every US lab has these capabilities or better behind the scenes and isn’t shipping them due to risk aversion, some of which comes from regulatory risk.” This is a plausible explanation. Major AI labs like OpenAI, Anthropic, and Google are likely subject to significant internal and external pressures, including concerns about safety, ethics, and potential misuse of their technologies. These concerns could lead to a more cautious approach to releasing products with advanced agentic capabilities.
Regulatory risk is also a significant factor. Governments around the world are grappling with how to regulate AI, and the uncertainty surrounding future regulations could make companies hesitant to release products that might be subject to restrictions or even bans. This “risk aversion” could explain why a smaller, more agile startup like Manus might be willing to push the boundaries further than established players.
It’s also possible that the major labs are focusing on different priorities. They might be more interested in developing the underlying models themselves, rather than building user-facing applications. Or they might be working on different types of agentic capabilities that are not yet ready for public release. Whatever the reason, the emergence of Manus highlights a potential opportunity for smaller companies to innovate in the application of AI, even if they don’t have the resources to develop their own foundational models.
Open Source Aspirations and the Emergence of OpenManus
The fact that Manus is built on existing LLMs suggests that its capabilities could potentially be replicated. This realization sparked a wave of anticipation among many users on X, with some expressing hope for an open-source version. The idea was that, if the core functionality of Manus relies on publicly available models and tools, it should be possible to create an open-source alternative that offers similar capabilities.
These hopes appear to have been answered relatively quickly. A group of developers on GitHub has already created an open-source alternative to Manus, aptly named “OpenManus.” This project is now publicly available on GitHub. The rapid development of OpenManus demonstrates the power of the open-source community to quickly respond to new technologies and create alternatives. It also highlights the potential for open-source projects to democratize access to advanced AI capabilities.
OpenManus is still in its early stages of development, and it remains to be seen whether it will be able to fully replicate the functionality of Manus. However, its existence is a significant development, and it could potentially serve as a platform for further innovation in the field of AI agents. It also provides a valuable alternative for users who are concerned about the closed-source nature of Manus or who want to contribute to the development of an open-source solution.
Critiques and Challenges Faced by Manus
Despite the positive reception, Manus has also encountered its share of criticism. Some users have reported that Manus took an excessive amount of time to complete tasks, and in some cases, failed to finish them altogether. Derya Unutmaz, a biomedical scientist, compared Manus to OpenAI’s Deep Research, noting that while the latter completed a task in 15 minutes, Manus AI failed after 50 minutes, getting stuck at step 18 out of 20. This comparison highlights a potential performance gap between Manus and existing solutions, at least in some specific cases.
Simon Smith, EVP of generative AI at Klick Health, attributed these issues to the possibility that Manus’s underlying model might not be as robust as OpenAI’s Deep Research. He further suggested that because Manus utilizes multiple models, it might require more time than Deep Research to generate a complete report. This is a reasonable hypothesis. The complexity of integrating multiple models and tools could introduce inefficiencies and potential points of failure.
Another user highlighted that Manus sometimes gets stuck during web searches, experiences “breaks in between” due to context issues on code-based tasks, and exhibits general slowness. These observations point to potential limitations in Manus’s ability to handle certain types of tasks, particularly those involving complex web interactions or code execution.
Some critics have also targeted Manus’s invite-only access approach, suggesting that invitations were primarily distributed to influencers on social media to generate hype. This criticism raises concerns about the fairness and transparency of Manus’s rollout strategy. While it’s common for companies to use influencer marketing, relying too heavily on it can create a perception of artificial hype and exclude potential users who might provide valuable feedback.
The Future of Manus and the Broader AI Landscape
It’s important to acknowledge that Manus is still in its early stages of development, and it is likely to undergo further refinement and improvement. The criticisms and challenges it has faced are not unusual for a new technology, and the development team will likely address them in future iterations. However, a crucial question remains: how long will it be before major players like OpenAI, Anthropic, or even Google introduce a more widely accessible version of what Manus currently offers?
The emergence of Manus serves as a compelling demonstration of the potential of AI agents and the value of creating user-friendly interfaces to unlock the capabilities of existing AI models. It shows that significant advancements can be made not just by developing new models, but also by finding innovative ways to apply existing ones. While challenges and criticisms exist, Manus represents a significant step forward in the evolution of AI-powered tools and their ability to tackle complex, real-world tasks.
The development of OpenManus further underscores the community’s interest in exploring and expanding upon the possibilities presented by this new approach to AI agents. The open-source nature of OpenManus could accelerate innovation and lead to even more sophisticated and accessible AI agents in the future. The future will likely see continued innovation and competition in this space, driving the development of even more sophisticated and accessible AI agents. The competition between closed-source and open-source approaches will be particularly interesting to watch, as it could have a significant impact on the accessibility and democratization of AI technology.