Biography of Xiao Hong and Development of Manus AI

Xiao Hong, Butterfly Effect Pte. Ltd., and the Rise of Manus: A Comprehensive Report

Introduction

The rapid evolution of artificial intelligence (AI) in the mid-2020s has produced several technological breakthroughs globally, but few have been as universally disruptive and internationally discussed as Manus. Recognized as one of the world’s first truly autonomous general AI agents capable of independent, real-world action—including writing and deploying code without ongoing human supervision—Manus’s impact has been both technical and symbolic, illustrating the global diffusion of AI innovation beyond traditional Silicon Valley boundaries^1,2. At the heart of this achievement sits Xiao Hong, the developer, entrepreneur, and strategist whose career trajectory from regional Chinese tech venture founder to international AI visionary frames the Manus story.

This report delivers an in-depth, structured analysis of Xiao Hong’s biography, the rise and transformation of Butterfly Effect Pte. Ltd., and the technical and societal significance of Manus as a landmark in global AI history. The narrative is supported by detailed discussion of investment history, organizational transitions, and the architecture and practical performance of the Manus agent, drawing from a comprehensive selection of diverse sources to ensure accuracy and depth. Key milestones are summarized in the table below, to be expanded in narrative in the subsequent sections.

Table: Xiao Hong’s Career Milestones and the Development of Manus

Year	Milestone	Associated Venture
1992	Born in China	—
~2014	Graduates from Huazhong University of Science and Technology (HUST)	—
2015	Founds Wuhan Nightingale Technology Co., Ltd.	Nightingale Technology
2019-2021	Raises multiple funding rounds for Nightingale	Nightingale Technology
2022	Founds Butterfly Effect Technology Group; launches Monica browser plugin	Butterfly Effect
Aug 2023	Incorporates Butterfly Effect Pte. Ltd. in Singapore	Butterfly Effect
Dec 2024	Reportedly exits as shareholder of Monica	Butterfly Effect / Monica
Mar 2025	Launches Manus AI via invite-only beta; scores high on GAIA benchmark	Butterfly Effect / Manus AI
Apr 2025	Raises $75M Series B funding led by Benchmark	Butterfly Effect
Jun 2025	Publicly announces full relocation of HQ to Singapore	Butterfly Effect

I. Xiao Hong: Biography and Entrepreneurial Ascendance

1. Early Life and Education

Xiao Hong (肖弘), widely known in English as “Red Xiao,” was born in 1992 in China—against the backdrop of dramatic economic reform and digital acceleration. He entered higher education at Huazhong University of Science and Technology (HUST), a top Chinese institution known for its rigorous engineering and computer science programs³. Xiao completed a degree in Software Engineering at HUST, likely around 2014, equipping him with the technical foundation and exposure to contemporary software engineering trends that would later underlie his ambitious entrepreneurial undertakings.

Emerging as a member of the cohort that experienced China’s meteoric tech growth in real time, Xiao’s environment was shaped by the rise of influential companies like Tencent, Alibaba, and Baidu—firms that would become not only technological inspirations but also, in some cases, crucial investors for his ventures³. By entering the workforce in the first half of the 2010s, Xiao accessed an unprecedented confluence of capital, technical talent, and globalizing ambition, intensifying his appetite for rapid-market product development.

2. Nightingale Technology: Foundation and Growth

Xiao’s formal entrepreneurial journey began shortly after university, when he founded Wuhan Nightingale Technology Co., Ltd. in 2015. The company’s raison d’être was the development of AI-powered productivity tools that could integrate seamlessly with existing business and social platforms—most notably WeChat, which had cemented its position as China’s dominant super-app^4,5.

Nightingale’s flagship products included the “Yi Ban Assistant” (易班小助手) and the “Wei Ban Assistant” (微班小助手), both of which rapidly captured enterprise market share. Functioning as AI- and automation-enhanced platforms for workplace productivity and messaging, these tools reached over two million B2B (enterprise) users and, according to broad estimates, indirectly served hundreds of millions of consumer users by improving B2B2C workflows^4,5.

The rapid growth and market resonance of these platforms enabled Nightingale to attract funding from a roster of premier investors—including Tencent Investment, ZhenFund, and other noted VCs. By 2019-2021, Nightingale had closed multiple venture rounds, reaching hundreds of millions of CNY in valuation and solidifying Xiao’s reputation as a product-oriented executive who could marry technical insight with business execution⁵.

3. Product Philosophy: Kitbashing and “Wrapper” Strategies

A core feature of Xiao’s approach, evident from the earliest Nightingale products, was his philosophy of “kitbashing”—the creative integration or wrapping of AI capabilities onto entrenched, widely used platforms, such as WeChat⁶. Rather than creating entirely standalone solutions, Xiao’s companies specialized in delivering value-adding layers that could rapidly deploy and scale by leveraging existing infrastructure and user habits. Whether through custom WeChat CRM plugins, official account editing tools, or browser-based intelligence overlays, Xiao’s projects reflected a pragmatic focus on user-centric integration and low initial friction.

This approach, which prioritized accessibility and instant utility over deep foundational model innovation (at least initially), would presage many of his subsequent moves in the global AI agent space.

4. Transition to Butterfly Effect and the Monica Plugin

By 2022, as global interest in large language models (LLMs) and AI agents intensified, Xiao Hong shifted focus to founding Butterfly Effect Technology Group. Butterfly Effect first operated as a Beijing-based and then Hong Kong-based entity, with subsequent expansion to Singapore. Its initial high-visibility product, Monica (monica.im), was released as an AI browser plugin and rapidly garnered more than one million users, primarily in overseas markets^3,7,8.

Monica exemplified Xiao’s kitbashing philosophy: it provided a flexible browser sidebar and content assistant integrating multiple major LLMs (Claude, GPT-4o, DeepSeek, Gemini) through a unified interface, enabling power users to translate, summarize, search, write, rewrite, and analyze content directly on any webpage. The plugin’s seamless extension model and strong user feedback mechanisms foreshadowed the strategic product thinking that would lead directly to Manus⁸.

5. Investor Relations and Exits

Throughout these ventures, Xiao demonstrated both charisma in attracting high-profile backing and prudence in timely exits. He reportedly exited as a shareholder of Monica by December 2024, possibly in anticipation of regulatory headwinds and to focus executive attention on international expansion and the more ambitious Manus project⁵. A similar pattern occurred in Nightingale, where Xiao had departed all shareholder positions prior to, or during, the client migration to the Butterfly Effect ecosystem, setting in motion a global-facing phase in his career.

II. Butterfly Effect Pte. Ltd.: Company Evolution and Internationalization

1. Founding and Early Focus

Butterfly Effect Pte. Ltd. was first incorporated as a Singapore-based company (UEN 202330764R) on 2 August 2023, with its origins stretching back to prior Beijing and Hong Kong entities dating to 2022^9,5. The company’s principal activities included the development of software applications and publishing, with an explicitly international orientation. The Singapore registration—mirrored by a Cayman Islands holding structure—established Butterfly Effect as both a regional and a global enterprise, aiming to act as a bridge between East Asian innovation and worldwide markets¹⁰.

Butterfly Effect’s early reputation had been built off the momentum of the Monica browser plugin, which itself was distinguished in the LLM plugin sector for its quick adoption, responsive iteration, and broad user appeal. The company’s ability to integrate Claude, OpenAI, DeepSeek, and Gemini models into a single, user-controlled sidebar not only offered operational flexibility but also provided the team experience in building robust middleware layers for AI orchestration—an architectural capability that would be dramatically scaled up in Manus^7,8.

2. Investment Rounds and Strategic Backing

Butterfly Effect benefited from a series of high-profile funding rounds. Two rounds in early 2025 raised more than $10 million USD, with initial support from ZhenFund (Xiao’s longtime angel investor), and subsequent funding from Sequoia China, Tencent, and entrepreneur Wang Huiwen¹¹. In April 2025, a milestone $75 million USD Series B round led by Benchmark, a leading Silicon Valley firm, propelled the company’s valuation to approximately $500 million, up from a previous $100 million just months earlier^12,13. These rounds included continuing participation from Tencent and ZhenFund, and, reportedly, HSG (formerly Sequoia China), exemplifying robust confidence from both Chinese and international investors in Butterfly Effect’s autonomous agent vision.

The pattern of investment and cross-border structuring reflected a calculated positioning to attract Western venture capital, navigate tightening regulatory controls on AI investment and hardware exports, and pursue rapid, capital-intensive development of global AI platforms⁹.

3. Relocation to Singapore: Geopolitical and Regulatory Motives

By June 2025, Butterfly Effect publicly announced a full relocation of its headquarters from China to Singapore. This move was motivated by intersecting strategic imperatives: the intensification of Sino-U.S. tech and semiconductor tensions; U.S. export controls on Nvidia and other advanced chips; and the increasing compliance burden on AI startups operating under both Chinese and global regulatory regimes^14,10.

Singapore’s neutral status in global geopolitics, business-friendly environment, and deep reservoir of technical talent made it an appealing alternative. By mid-2025, Butterfly Effect had initiated significant downsizing of its operations in China (reducing a 120-person team to a ~40-person technical core), shifting hiring to Singapore to focus on high-skill AI and software engineers with internationally competitive salaries ($8,000-$18,000/month)¹⁰.

This headquartering strategy mirrored those of other aspiring global AI firms like HeyGen and WIZ.AI, and provided Butterfly Effect with direct access to U.S., Japanese, and Middle Eastern markets, as well as a cleaner compliance profile for investors concerned about the patchwork of global tech regulations¹⁴.

4. Governance and Leadership

Alongside Xiao Hong, the core Butterfly Effect leadership includes Ji Yichao (Chief Scientist), Zhang Tao (Head of Product), and other industry veterans from Tencent, ByteDance, and Peak Labs^5,4. This multi-founder structure disseminated technical, product, and operational responsibilities, positioning the company to blend technical credibility with business agility. Zhang Tao, for instance, was central in the company's high-profile public announcements at the June 2025 SuperAI conference in Singapore.

5. Strategic Focus: From Foundation Models to Application Layer

A defining trait of Butterfly Effect’s business model is its explicit pivot away from building foundational large language models (LLMs) due to the hardware resource bottlenecks and U.S. export controls, choosing instead to focus expertise in the application and agentic orchestration layer¹⁰. By developing agentic software platforms capable of integrating multiple best-in-class models, Butterfly Effect reduced its dependence on one nation’s hardware ecosystem, increased operational flexibility, and aligned better with the trajectory of Western AI investment (which increasingly prioritized application over research-level model innovation).

III. Manus AI: Launch, Architecture, Performance, and Reception

1. Genesis and Naming

Manus AI was officially launched via invite-only closed beta on March 6, 2025—its name derived from the Latin for “hand,” evoking the classic engineering motto “Mens et Manus” (mind and hand), a nod to the combination of reason and practical execution that defines the agent’s philosophy³. Xiao Hong presented Manus not simply as a chatbot or conversation assistant, but as a “general AI agent” whose mandate was to turn high-level user intent into real-world action, closing the loop from cognition to deployment^15,16.

The connection to Monica is explicit: Manus is built by an evolved variant of the Monica team, extending the product philosophy of integrated, user-controlled AI workflows by moving from browser overlays to a full, autonomous agentic system capable of independent planning, coding, and web manipulation.

2. Technical Architecture

a. Multi-Agent Orchestration

At the heart of Manus’s technical architecture is a multi-agent system, in which specialized sub-agents coordinate iteratively under the oversight of a high-level “executor” or “planner” agent. The core system is built atop large language models such as Anthropic’s Claude 3.7 Sonnet and fine-tuned versions of Alibaba’s Qwen models, depending on the deployment context^17,21.

Key Subsystems:

Planner Agent: Breaks user requests into structured action plans and sub-tasks, managing overall workflow sequence.
Executor/Knowledge Agents: Perform step-by-step execution (shell commands, browser manipulation, API calls), and perform real-time knowledge retrieval and fact-checking.
Verifier/Quality Agents: Check result correctness and iterate further if needed, correcting errors or rerunning steps.
Specialized Toolkits: There are 29 (as per some internal architecture leaks) specialized toolkits for tasks such as web interaction, file system control, data extraction, and code deployment¹⁸.

b. Secure Linux Sandbox/Cloud Workspace

Manus operates in a cloud-based Linux sandbox environment, allowing the orchestration of multi-step tasks such as:

Writing and running shell scripts
Installing, editing, and deploying code artifacts (websites, dashboards, applications)
Managing files, folders, and intermediate outputs
Running browser-based automation (form filling, scraping, interaction with web services)

All of this is performed securely and asynchronously, with each session isolated from others to protect privacy and prevent cross-user access^19,1.

c. Persistent, Asynchronous Execution

Unlike conventional chatbots, Manus can continue working on tasks after the user disconnects, enabling “fire-and-forget” workflows for multi-hour, or even multi-day, research and deployment operations—vital for business automation and batch processing^20,21,22.

d. Real-Time Workflow Visualization

A distinctive Manus feature is the “Manus’s Computer” virtualization window, where users can observe, intervene, and interact with the running agent as it browses websites, runs scripts, and generates output. This transparency not only aids trust but also lets users take over in case of stuck flows (e.g., CAPTCHAs), a problem most AI agents struggle with¹⁹.

3. Benchmark Performance and Technical Comparison

a. GAIA Benchmark Dominance

Manus debuted with best-in-class scores on the GAIA benchmark (General AI Assistant benchmark), a state-of-the-art testbed designed by Meta, Hugging Face, and the AutoGPT team. The GAIA evaluates real-world AI agent task execution—requiring multi-modal sensemaking, tool use, and reasoning—not just conversational coherence.

Level 1 (Basic tasks): 86.5% (vs. OpenAI Deep Research 74.3%, previous SOTA 67.9%)
Level 2 (Intermediate tasks): 70.1% (vs. OpenAI Deep Research 69.1%)
Level 3 (Complex tasks): 57.7% (vs. OpenAI Deep Research 47.6%, previous SOTA 42.3%)

These scores positioned Manus above even best-in-class open-source and proprietary Western agents in every measure and underscored its capacity for persistent real-world action in complex digital environments^19,20.

b. Functionality Comparison to Peers

Manus’s “bundle and integrate” kitbashing approach, using a wrapper philosophy around the best LLMs and browser automation stacks, gave it early-mover practical advantages, even if its base model was not always proprietary or novel^21,8.

4. Use Cases and Demonstrations

Coding and Deployment: Manus has demonstrated the capability to write multi-language code, debug, test, and deploy entire web servers, analytical dashboards, and user-facing tools—all based on single high-level prompts^19,21.
Data Analysis and Visualization: It automates the ingestion of large, unstructured datasets, performs comparative analysis, and outputs both textual and graphical reports, often creating shareable public outputs (e.g., dashboards, presentations, Excel/Word files)^23,22.
Research, Planning, and Business Process Automation: Manus’s cloud-executed workflows enable use cases from complex travel and event planning, to business analysis (SEO audits, supplier sourcing), to screening resumes and summarizing large-scale research, all without continuous user engagement^20,15.
Customer Reception and Feedback: Initial access codes fetched upwards of $14,000 USD on black markets, and Manus’s Discord community reached over 180,000 members within a month of launch²⁴. Despite some early criticisms of slow execution (for very complex chains), echoing limitations reported in media tests (system crashes, looping errors on ill-posed tasks), the system was widely praised for intuitive interface, real-time transparency, and “intern-like” versatility^18,23.

5. Critical Reception and Media Analysis

Global coverage of Manus was intense. Western business and tech press (Forbes, VentureBeat, Newsweek, MIT Technology Review) declared it a “second DeepSeek moment”—i.e., not just matching Western innovation but redefining assumptions about the locus of practical AI progress^15,25. Chinese and Southeast Asian outlets celebrated its ability to break free from China-U.S. tech bottlenecks by leveraging international structuring, while developer and open-source communities closely scrutinized its architecture and source code for confirmation of true autonomy versus sophisticated tool orchestration^22,3.

6. Monetization and Growth Dynamics

Access during the initial phase was invite-only, creating high pricing for codes and fostering a sense of exclusivity. By May 2025, the initial user boom (peaking at 20 million MAU) had contracted to roughly 10 million active monthly users, a common pattern in beta launches challenged by new domestic Chinese competitors (ByteDance, Baidu) and the “war of a hundred models” phenomenon within the Chinese AI ecosystem¹⁰. Still, continued infusions of capital and a strongly loyal user core, especially in developer and business automation verticals, positioned Manus as a long-term platform play rather than a hype-driven one-off.

IV. Xiao Hong’s Impact, Product Philosophy, and Future Vision

1. Kitbashing and Technical Integration Strategy

Xiao’s enduring influence in the AI sector has been his approach to value creation through expert integration—rapidly packaging best-in-class LLMs, open-source frameworks, and custom toolkits rather than sinking years and capital solely into model research. This “kitbashing” or “wrapper” approach, as referenced by both Western and Chinese media, enabled speed of iteration, nimble response to regulatory headwinds, and scalable incremental upgrades, while also exposing the product to criticism for lack of foundational innovation^17,2.

Nonetheless, in the context of real-world enterprise and developer productivity, Xiao correctly anticipated user demand for practical, outcome-oriented platforms, rather than simply state-of-the-art conversational models. By focusing on orchestration, tool chaining, and workflow transparency, he put Butterfly Effect and Manus ahead in the emerging battle for AI agent utility.

2. Open-Source Commitment and Ecosystem Ambitions

Xiao Hong has publically announced plans to open-source parts of Manus’s model and toolchain—an ambition intended to further the democratization of agentic AI and reduce global dependency on proprietary, paywalled Western cloud APIs^3,26. This move is calculated to grow developer engagement, quickly expand the Manus ecosystem, and preempt competitive erosion by global open-source projects such as AutoGPT, OpenManus, and OWL.

The OpenManus project, started as a community-driven response to Manus’s high beta exclusivity, is now developing rapidly as an open-source alternative, with dynamic tool generation and modular architecture for AI automation that can be self-hosted, adapted, and benchmarked independently²⁶. Xiao’s support of these efforts aims to position Manus less as a single product and more as a movement and platform around autonomous, transparent, globally accessible AI.

3. Ethical and Societal Considerations

Xiao has been outspoken about the ethical complexities of AI agent deployment—from privacy in cloud-based orchestration, to the risk of autonomy mistakes (wrong financial trades, security breaches), to questions of ultimate responsibility when AI acts without human supervision^1,15. In contrast with some Western AI leaders who shy away from fully autonomous operation, Xiao’s model assumes a future in which AI agents become not just assistants but independent actors in digital and physical infrastructure.

He and his team have emphasized the need for detailed verification agents, transparent toolchains, and user-in-the-loop override features, and they have worked on compliance with data protection standards (such as GDPR for European markets). As more users and enterprises automate “mission-critical” processes with agents like Manus, these design and governance issues will only grow in importance.

4. Personal Brand and Leadership

Xiao Hong projects a leadership style that is both technically engaged and strategically dispassionate—unafraid to exit ventures when the time is right, to dissolve legacy operations for more competitive international structures, and to refocus on scalability and resilience at the application layer. His migration of Butterfly Effect from China to Singapore, his willingness to accept U.S. VC capital amid regulatory scrutiny, and his embrace of open-source all signal an adaptive, world-facing approach uncommon in the often inward-focused Chinese tech sector.

Xiao’s “less structure, more intelligence” philosophy encapsulates the Manus ethos: reducing barriers to thought-to-action translation, prioritizing flexible data pipelines and modularity above single-model worship, and—at the broadest level—redefining the meaning of “intelligence” in software as the ability to autonomously deliver, not just to answer.

V. Strategic Challenges and Future Trajectory

1. Market Pressures and Global Competition

While Manus’s early GAIA benchmark dominance and strong capital inflows secure it a first-mover advantage, competition is accelerating rapidly. Domestic Chinese players like ByteDance and Baidu have upgraded their agentic toolchains, and international rivals like OpenAI, Anthropic, Google DeepMind, and independent open-source agents are iterating rapidly on autonomous orchestration architectures^23,17,2. Manus’s user base contracted after the honeymoon period, reflecting both typical beta-cycle hype and the reality that reliable, scalable, enterprise-grade agent deployment remains challenging.

2. Technical Challenges

Persistent frustrations around execution speed, reliability (e.g., browser CAPTCHAs, website login requirements), and error recovery have been noted by media and early testers. Manus’s reliance on cloud compute, and its exposure to shifting policies and API costs of underlying model providers (e.g., Claude, Qwen), create ongoing bottlenecks for both cost and deterministic operation^23,18. Although the modular, wrapper model enables composability and rapid feature extension, it also exposes Manus to vulnerabilities if core API providers change pricing or access models.

3. Open vs. Proprietary Approaches

With the emergence of open-source alternatives (OpenManus, OWL, ANUS), the manuscript’s original advantage as the only practical autonomous agent is quickly dissolving. Xiao’s commitment to open-source is not only strategic (keeping developers loyal and engaged) but also necessary to avoid lock-in backlash and ensure momentum as the field shifts to collaborative innovation rather than walled-garden solutions¹⁸.

4. Societal and Economic Impact

As Manus and peer agents begin to automate the types of skilled knowledge work long considered safe from AI disruption—coding, business analysis, research, even planning and negotiation—the broader consequences for labor, ethics, and regulation are vast. Xiao’s work, while focused on product and technical leadership, will increasingly place him at the center of debates on responsibility, agency, and the future structure of the global workforce^1,15.

VI. Conclusion

Xiao Hong’s journey from a software engineering student at HUST to the global face of agentic AI development reflects the logic and ambition of 2020s-era technology entrepreneurship: rapid cycling through product generations, fearless structural reinventions, and relentless prioritization of user-centered, outcome-driven innovation. The development of Manus, under his direction, marks a genuine leap in AI’s practical autonomy and global accessibility, setting new standards for what intelligent software can deliver in the real world^1,2.

As Butterfly Effect cements its position in Singapore and prepares for further Western and Asian expansion, the global AI community will continue to watch the evolution of Manus not only as a platform but as a template for future AI agency. Xiao’s highly iterative, open, and solution-focused methodology—emphatically not restricted by national, regulatory, or even technical orthodoxy—captures the essence of a new era: where the boundary shifts from “what can AI say?” to “what can AI actually do, unsupervised?” The economic, social, and ethical ripples of that paradigm shift are only beginning to be felt.

End of Report