I tried Manus: it's not the tech breakthrough you think it is
Manus's business philosophy: time to think beyond just tech
Since Manus launched, it has created quite a buzz, and all AI enthusiasts are eager to get their hands on an invitation code to actually use it. The AI agent is widely considered to be the next frontier for AI startups and investment, and Manus claims that it's the first-ever general-purpose AI agent. If it’s as good as it advertises, we may be seeing a significant step toward AGI. Many had hoped that Manus could be another "Deepseek" moment for China's tech scene and even serve as a booster for the recent tech rally. However, since its launch, it has been incredibly difficult to obtain an invitation code to test it out.
Last week, we published a post questioning whether Manus is a genuine technological breakthrough or just a marketing gimmick. However, some existing analyses and studies suggest that Manus is still far from being a second "Deepseek moment."
There’s also skepticism about how groundbreaking Manus truly is. Some Chinese teams have already built open-source versions that replicate Manus in just a few days. As a result, many argue that Manus is just a wrapper around existing large language models, with no real barrier to entry.
Over the weekend, I also created a demo video showcasing Manus in work-related scenarios, and my conclusion remains similar. At this stage, I don’t see much technological innovation at the code level with Manus.
(You can check out my review of Manus here)
But does that mean Manus is simply a marketing scam with no value? I wouldn’t jump to conclusions just yet. I believe the team is intentionally working with alpha users. Building a general-purpose AI agent is an ambitious goal, and the team likely needs to narrow their focus to key use cases with the greatest potential for commercialization. (Although not being able to get an invitation code easily does feel annoying and a bit fishy.)
Both Deepseek and Manus founders are Chinese-educated talents, but Manus's founder, Xiao Hong, has a very different philosophy from Deepseek's founder, Liang Wenfeng. While Liang may seem like the bigger star because of his focus on foundational technological breakthroughs, Xiao is a serial entrepreneur with years of experience in the field. Most of his past ventures have centered on integrating existing AI models into practical applications. For example, he previously launched Monica AI, an AI-powered web browser that gained over 3 million active users worldwide. This time, his team has again focused on international markets since day 1 to avoid direct competition with China’s tech giants.
In today’s newsletter, I’m translating an article originally published by The Paper that dives into the story behind the Manus team. The point isn’t just to learn about Manus, but to understand how many Chinese companies approach their global strategies. Instead of focusing exclusively on creating technological leadership, these companies often specialize in "repackaging" existing tech to create commercially viable products. Their story suggests that if one can master "repackaging"—even if it's just building a wrapper around existing models, which appear to have no obvious technological barrier—it can still lead to commercially successful products.
After the translated article, I’ll follow up with my thoughts on the business implications of Manus for our Baiguan readers. Let's dive in.
Behind Manus: How a Chinese team mastered the art of 'repackaging' to win big
The AI world exploded last night with the launch of Manus, a cutting-edge AI agent that's taking the tech community by storm. At this moment, every AI enthusiast is scrambling for an invite to this groundbreaking product.
What is Manus？
Hyan from the Manus team shared on Superlinear Academy, an online learning platform and community dedicated to AI enthusiasts, that Manus is the world's first general-purpose AI agent capable of independently executing a wide array of complex and diverse tasks. Whether you require sophisticated market research, tedious bulk document processing, personalized travel planning, or professional data analysis, Manus can autonomously strategize and execute tasks within its virtual environment. It dynamically leverages various tools—writing and executing code, intelligently browsing the web, and interacting seamlessly with web applications—to deliver complete results rather than merely providing suggestions or answers.
JAZZYEAR, a prominent tech think tank, classifies AI agent capabilities into four levels, L1 through L4. Based on Manus's product demonstrations, Manus has nearly reached Level 4 autonomy—the apex of AI workflow automation—heralding a revolutionary shift in personal productivity.
But it's not just the remarkable product that's catching attention. Equally intriguing is the Chinese team behind Manus: Monica.im, led by founder Xiao Hong, a serial entrepreneur. In 2023, Monica.im launched Monica, an AI-powered browser assistant plugin. Before that, Xiao had already made waves developing popular WeChat enterprise SaaS tools—Weiban Assistant and Yiban Assistant—that became essential tools for many media professionals and independent creators.
Rather than developing their own foundational AI models, the Monica team strategically builds products by 'repackaging' existing large language models. At a time when industry attention is largely fixated on foundational AI developments from giants like OpenAI and DeepSeek, Monica has quietly and effectively navigated the application layer, turning the controversial practice of 'extreme repackaging' into their secret weapon.
While some in the industry dismiss repackaging as superficial or inferior, Monica.im is clearly demonstrating that pushing repackaging to its fullest potential can indeed lead to spectacular success.
From Yiban Assistant to Manus
The founder, Xiao Hong, a 2015 software engineering graduate of Huazhong University of Science and Technology, is a seasoned tech entrepreneur with over a decade of experience. During his university years, Xiao joined the Lianchuang team, actively engaging in entrepreneurial projects and leading the launch of several campus innovations such as the Volunteer Application Assistant, MieMie, and Circle Market, laying a solid foundation for his future entrepreneurial journey.
In 2015, after graduation, Xiao Hong founded Nightingale Technology, launching enterprise-focused WeChat SaaS tools—Weiban Assistant and Yiban Assistant. The company secured investments totaling hundreds of millions of RMB from renowned institutional investors, including Tencent and ZhenFund. These tools served over two million enterprise users (B2B) and helped businesses reach hundreds of millions of consumers (B2C). Ultimately, Xiao sold Nightingale Technology to a unicorn company in 2020. This entrepreneurial phase validated Xiao’s business strategy of building vertical tools atop powerful platform ecosystems.
In 2022, Xiao Hong foresaw the transformative potential of the AI revolution triggered by ChatGPT and founded Butterfly Effect, launching the AI-powered browser plugin Monica, strategically targeting international markets.
Monica is an all-in-one, browser-based AI tool offering everyday AI functionalities through a purely AI-driven approach. It integrates popular large language models, enabling users to engage in chats, translations, content processing, and graphic creation anytime, anywhere. Xiao positions Monica as a versatile AI assistant that helps users access, process, and store information directly via a browser extension. Looking ahead, Monica aims to become a highly personalized assistant tailored to individual users.
Xiao’s strategic decisions were informed by two key insights: firstly, to avoid intense domestic competition with tech giants like Baidu and Alibaba by targeting Western markets; and secondly, to redefine the value of technological repackaging. He argues that AI application companies should emulate consumer electronics companies, such as Apple, by integrating existing large-model APIs to create differentiated user experiences rather than pursuing foundational technological breakthroughs.
At the 2023 Jazzyear Gravity Conference, Liu Yuan, a partner at ZhenFund, commented on Monica’s international market strategy: “We've observed that domestic startups predominantly focus on B2B, while many companies going overseas target the B2C market. The regulatory uncertainties surrounding China's domestic B2C market have made entrepreneurs cautious, but Chinese entrepreneurs' capability to succeed in overseas B2C markets has significantly improved compared to five or six years ago. For example, Monica.im and HeyGen—both part of our investment portfolio—quickly attracted significant international attention. Furthermore, companies building their own foundational AI models we invested in also have ambitious international expansion plans. This marks a significant shift compared to the previous generation of AI entrepreneurs.”
In 2023, Monica accelerated its growth by acquiring "ChatGPT for Google," a popular product developed by independent developers. At the time of acquisition, ChatGPT for Google already had three million users. Integrating this plugin allowed Monica to rapidly expand its product portfolio, validating strong market demand for an integrated suite of AI tools.
Currently, Monica boasts approximately four million users, while ChatGPT for Google has around three million users, totaling more than seven million users combined. This user scale and breadth of functionalities position Monica among the leading companies in the AI Chrome plugin market.
Baiguan:
Innovation and Entrepreneurship HUST is the official WeChat account managed by the Innovation and Entrepreneurship Education Office of Huazhong University of Science and Technology (HUST), covering educational reforms, research achievements, entrepreneurial practices, project incubation, and investment updates related to innovation and entrepreneurship by students, faculty, and alumni.
In October 2024, at a startup event jointly organized by HUST and ZhenFund, Xiao emphasized the importance of entrepreneurs focusing on solving concrete technical problems rather than indulging in grandiose ideas. He also stressed the critical importance of making correct decisions during favorable times.
On March 6, 2025, the Monica team made headlines again by launching Manus, an AI agent that instantly captivated the technology community. According to the official website, Manus surpasses OpenAI Deep Research in capabilities.
Using GAIA, a benchmark for assessing general AI assistants in solving real-world tasks, Manus achieved new state-of-the-art (SOTA) performance across all three difficulty levels.
Less structure more intelligence
Both Monica and Manus exemplify the art of "extreme repackaging."
As an AI application-level startup, Xiao Hong does not aim to develop proprietary large language models. Instead, his company dynamically integrates multiple established models—such as GPT-4, Claude 3, and Gemini—to enhance overall capabilities, a strategy known as "repackaging." However, Manus has not yet disclosed the specific foundational models it leverages, though it likely employs a combination of several different models.
Xiao believes repackaging itself isn't problematic; the key question is whether it effectively solves real user problems. He draws an analogy between foundational AI model providers and chip manufacturers, while AI application companies parallel consumer electronics firms. Chip makers focus on creating better, more cost-effective chips, whereas consumer electronics brands prioritize branding, distribution, and differentiated user experiences. The fundamental mission of consumer electronics companies is to deliver tangible user value, thereby achieving a sustainable business model.
Thus, repackaging can be viewed as a strategic approach to integrating and optimizing resources. As long as it fulfills user needs and provides a superior user experience, repackaging represents a successful product strategy.
Xiao believes extreme repackaging effectively bridges technology-product fit (TPF) and product-market fit (PMF), ultimately driving user value.
The Monica team closely integrates large AI models with user demands to provide efficient and convenient AI-powered services. For example, Monica tailors its interactions and functionalities specifically to platforms like YouTube, Twitter, Gmail, and The Information, adapting to each platform's unique user needs. This deep understanding and precise fulfillment of user requirements have significantly driven Monica’s international success, doubling its user base while maintaining impressive profitability.
However, not everyone agrees with this "repackaging" approach.
At a recent GDC conference in Shanghai, Liu Hua, Vice President of MiniMax, argued that product innovation should prioritize foundational model upgrades rather than superficial enhancements.
Citing DeepSeek and MiniMax’s Talkie as examples, Liu highlighted that DeepSeek amassed over 100 million users within two weeks, faster than even ChatGPT, primarily due to the substantial capability upgrade of its foundational V3/R1 model. Similarly, Talkie significantly surpassed its main overseas competitor, Character.ai, after Google's acquisition of Character.ai’s foundational model team, because MiniMax's integration of model and application layers facilitated more effective iterative upgrades.
In summary, MiniMax argues that enhancing foundational AI models matters significantly more than superficial product refinements. However, whether or not one builds proprietary models is ultimately just a technological approach. Xiao Hong emphasizes that the core value of any product lies in solving real user problems.
The Monica team consistently prioritizes user needs throughout their product development. By encapsulating complex technical processes in the cloud, Monica delivers a simple and intuitive user interface. For instance, Monica enables users to perform advanced AI tasks without needing to understand technical details such as Python dependency management or API key configurations. This approach substantially lowers the entry barrier, allowing users to effortlessly enjoy the convenience brought by AI.
After the product launch, Hyan from the Manus team introduced their product philosophy at Superlinear Academy. He explained that the team firmly believes in the principle of "less structure, more intelligence," emphasizing that when the data quality is sufficiently high, the AI models powerful enough, the architecture flexible, and the engineering robust, advanced capabilities such as computer use, deep research, and coding agents naturally emerge as inherent abilities rather than mere features. He suggested delegating routine tasks to Manus, allowing users to focus on more creative work, while conveniently monitoring the task progress through mobile devices. By the time users return, Manus will have delivered satisfactory results.
In 2025, widely dubbed the "Year of AI Agents," a productivity revolution has officially begun.
Closing thoughts: it's time to think beyond just tech
Although general-purpose AI sounds comprehensive and one-fits-all, I'm sure the building process is just the opposite - it will take a lot of detours and problem-fixing before it can really work. During this process, a lot of the "dirty work" isn't that intelligent or smart; it's just manual human effort that needs time. So what we're seeing right now is an MVP product that illustrates the concept, core functionality, and vision of what the Manus team is trying to build for the future. And as a beta version and prototype of tomorrow's concept, I think it's good enough.
Since Baiguan focuses on China investment and business insights, I want to share a few thoughts on how Manus could affect China's tech investment scene. Manus team just announced a collaboration with Alibaba's Qwen team, and they plan to work together on open-source LLM models. Right now, I don't think it will materially impact the revenue of Alibaba or other Chinese companies with AI-related services, because not that many requests are being made from Manus AI yet. But I think it's a good step toward building downstream AI applications.
The long-term implication is simple: the more people use downstream AI applications, the more upstream cloud services, data centers, and large model companies can eventually justify their valuations. In the short term, I think it's a nice sentiment boost for the recent rally in Chinese tech stocks, but objectively speaking, I certainly don't think it's another DeepSeek moment just yet.
The product is still in a very early stage. It wants to be fully autonomous, but right now it still needs lots of human intervention if you want it to actually replace human labor in work settings. For personal learning scenarios, I think it's an amazing tool that can really help you get ahead and complete knowledge graphs for things you don't already know how to do. But from a commercial perspective, I'm not sure how many individual users would actually use it regularly if the service is paid.
Therefore, while Deepseek feels like a 雪中送炭(timely help in a crisis) moment that gave Chinese equities a much-needed boost, Manus is more like a 锦上添花(cherry on top) moment— a nice sentiment booster, but I wouldn’t expect a rally similar to the one we saw before.
Regarding how groundbreaking Manus really is, my take is that, yes, at this stage, it may just be a wrapper around existing models, and it's true that there’s no significant barrier to replicating what Manus can do at the code level. But making an application that truly delivers working solutions requires tons of trials, errors, and accumulated use cases. There's also marketing, cost management, funding, company operations, talent acquisition - all aspects that matter in this competition. Those are all barriers and assets a company can leverage in the long run.
Given the broad scope of problems Manus is trying to tackle, I focus more on how the ROI will look for the Manus team, rather than how much of a technological breakthrough it is. Does it rely on ads like search engines? Or charge a monthly subscription fee? Can it truly integrate into daily workflows and provide practical solutions, beyond the "fancy" demos that show making a video game for someone who doesn’t know how to code—use cases that look impressive but may not be persuasive enough for someone to actually pay for? These are questions I've been thinking about but don't have answers to just yet. I am happy to discuss with you in the comment section below.
