The Rise of Everyone's AI Army
How Skills.sh and Clowdbot Signal a New Era
It is nearly midnight in a small apartment in an eastern Chinese city, and Liu Shiyang is still working. The young journalist, two years out of school, sits at his cluttered desk with a cold cup of tea beside his laptop. Tomorrow’s deadline looms, and he has three hours of interview recordings to process. But instead of the familiar dread, Liu feels something closer to curiosity. A WeChat tutorial is open in one browser tab, a configuration interface in another. Following the step-by-step guide, he is setting up something that would have seemed like science fiction to his professors: an AI system to transcribe his audio files, identify key quotes, cross-reference facts against public databases, and draft preliminary story outlines. The whole setup takes him about forty minutes.
What Liu Shiyang is doing in his late-night work session might appear to be simply adopting a new productivity tool. But look closer, and you will see something more significant unfolding. His quiet evening of technical tinkering connects to a much larger wave sweeping across the technology landscape. Thousands of miles away, in Silicon Valley and developer communities worldwide, two projects called Skills.sh and Clowdbot have recently sparked intense excitement. The connection between Liu’s tutorial and these viral phenomena is not coincidental. They are both manifestations of the same profound shift: we are entering what might be called the Age of Agents.
From “Conversation” to “Construction”: A Paradigm Shift in AI Applications
In early 2026, a personal AI assistant called Clowdbot (later renamed Moltbot) became a viral sensation, attracting attention from TechCrunch and sparking heated discussions across technology communities and social media platforms. But what made people truly excited was not what Clowdbot could accomplish in terms of tasks completed or problems solved. Instead, the enthusiasm stemmed from what it represented: a demonstration that ordinary people, not just programmers and AI researchers, could assemble their own multi-agent systems. The viral spread of Clowdbot was not about a product going viral. It was about a possibility going viral.
This possibility was made concrete by Skills.sh, a marketplace and directory for AI agent capabilities that Anthropic has been supporting. Before Skills.sh, anyone wanting to extend their AI assistant’s capabilities faced three significant hurdles. First, there was the discovery problem: how do you even know what skills exist and what your AI agent could potentially do? Second, there was the installation problem: even if you found a useful skill, configuring it often required technical expertise that most people lacked. Third, there was the sharing problem: if you created something useful, there was no standardized way to share it with others. Skills.sh addresses all three. It provides a one-stop directory where users can browse and search available skills. It offers one-click installation that eliminates complex configuration. And it establishes a standardized format called SKILL.md that follows the principle of “write once, shared everywhere.” A skill written for one Claude integration works seamlessly in other Claude-powered applications.
These developments represent the second stage in what has been a progressive democratization of AI technology. In the first stage, which began with ChatGPT’s public release, large language models became accessible to everyone. Suddenly, anyone with an internet connection could converse with sophisticated AI. Now, in the second stage, tools like Skills.sh and frameworks like those powering Clowdbot are enabling something more profound: ordinary users can become builders of multi-agent systems.
The shift to the second stage represents a fundamental change in the relationship between humans and AI. Consider the contrast. In conversational AI, the dynamic is question and response: users ask, the AI answers. In agent-based AI, the dynamic is goal and execution: users define objectives, and AI agents take autonomous action to achieve them. The output changes too, from generated content to completed tasks. Most importantly, the user’s role transforms from consumer to architect. And the barrier to creation drops dramatically: instead of needing programming skills, users can now assemble pre-built components like building with modular blocks.
Open Protocols and the Promise of Equal Opportunity
How has this third stage of democratization become possible? The answer lies in the emergence of open protocols and thriving open-source ecosystems. At the foundation is the Model Context Protocol, or MCP, which Anthropic released as a completely open standard under the MIT license. Think of MCP as a USB-C port for AI applications. Just as USB-C provides a universal interface that allows different devices to connect and communicate, MCP standardizes how AI assistants connect to external data sources and tools. This standardization is critical because it means developers can build integrations once and have them work across different AI platforms, rather than recreating the same functionality for every possible AI system.
The open-source ecosystem that has emerged around agent frameworks is remarkably vibrant. Microsoft’s AutoGen has attracted over 54,000 stars on GitHub and more than 550 contributors from around the world. LangGraph, CrewAI, and other major frameworks have all adopted the MIT open-source license, meaning anyone anywhere can use them freely. The adoption numbers are striking: according to CrewAI’s own data, 60 percent of Fortune 500 companies now use their platform, with over 450 million agent workflows running each month. These are not experimental toys but production systems delivering measurable business value. DocuSign reports that their AI agents reduced lead response time by 75 percent. General Assembly achieved a 90 percent reduction in development time for certain workflows. PwC saw code generation accuracy improve from 10 percent to 70 percent.
For observers in the Global South, there is another crucial piece to this story: the emergence of cost-competitive alternatives. DeepSeek, a Chinese AI company based in Hangzhou, has positioned itself as a provider of highly capable models at significantly lower prices than international competitors. DeepSeek’s strategy is not to claim the top spot on every benchmark but to offer practical, affordable AI that more people and organizations can actually use. For countries and institutions where budget constraints are real, this approach opens doors that might otherwise remain closed. The resource threshold for participating in the AI agent revolution has shifted from requiring massive capital investment to accessible, pay-as-you-go API calls.
China’s Pattern of Following and Innovating
Within this open ecosystem, how do latecomers find their footing? China’s experience offers an instructive case study that carries relevance far beyond its borders. Chinese enterprises have adopted what might be described as a “follow and innovate” pattern: in foundational technologies and protocol standards, they follow international developments; in application scenarios and local adaptation, they pursue distinctive innovations.
At the protocol layer, Chinese companies have moved quickly to support MCP, integrating it into their platforms to ensure compatibility with the broader ecosystem. At the framework layer, many rely on the same international open-source tools that developers worldwide use, including AutoGen, LangGraph, and similar projects. There is a clear-eyed recognition here: at this moment in the technology’s development, Chinese enterprises are primarily adopters of standards rather than setters of standards. This is not a source of shame but a pragmatic starting point.
Yet within this following posture, genuine innovation flourishes at the application and localization layers. Consider ByteDance’s Coze platform (known as Kouzi in Chinese). Leveraging its massive user base across Douyin (the Chinese version of TikTok) and the enterprise collaboration tool Feishu (known internationally as Lark), Coze has developed sophisticated workflow orchestration capabilities tailored to how Chinese users actually work. Huawei’s HMAF (HarmonyOS Multi-Agent Framework), announced at the 2025 Huawei Developer Conference, takes a different approach. It integrates deeply with the HarmonyOS operating system, enabling agents to coordinate across the full spectrum of Huawei devices, from smartphones to smart home appliances to vehicles. Huawei has articulated its own conceptual framework called ACT, standing for Awareness, Coordination, and Task, describing how agents perceive user intent, coordinate their actions, and execute concrete operations. This is not merely copying Western products but adapting the agent paradigm to the distinctive characteristics of the Chinese technology ecosystem.
The strategic lesson here extends beyond China. It suggests that latecomers to any technological wave need not achieve originality at every layer. What matters is finding the layers where differentiation is possible and focusing innovation there. In an era of open protocols and shared infrastructure, the barriers to participation have lowered. Those who can identify distinctive local needs and build solutions tailored to those needs can carve out meaningful positions even without controlling the foundational technologies.
A Window of Opportunity for the Global South
For readers observing from countries across the Global South, the emergence of the agent era presents a rare window of opportunity. Unlike previous waves of technological change, where participation often required massive infrastructure investments or access to proprietary technologies controlled by a handful of corporations, the agent revolution is building on foundations of openness.
The MCP protocol is not locked behind licensing fees. The major agent frameworks are not restricted to developers in wealthy nations. The skills being shared on platforms like Skills.sh are accessible to anyone with an internet connection. Low-cost API providers like DeepSeek make advanced AI capabilities available without requiring the kind of capital that only well-funded Silicon Valley startups or major corporations could previously afford. This is not to minimize the real challenges that remain: infrastructure gaps, talent shortages, and the continued concentration of cutting-edge research in a handful of countries are all genuine constraints. But the barriers are lower than they have ever been.
The path forward for Global South participants is not to attempt to compete at the foundational model layer, where the resource requirements remain enormous. Instead, the opportunity lies in the application layer, where understanding local contexts, languages, and needs provides a genuine competitive advantage. Consider the possibilities: agricultural agents that understand the specific crops, soil conditions, and market structures of particular regions; educational assistants that work effectively with local curricula and learning styles; government service agents tailored to the specific bureaucratic processes and citizen needs of different nations. These are not applications that companies in San Francisco or Shenzhen will build, because they require deep local knowledge that distant developers simply do not possess.
The recommended path follows a clear progression: start by using existing agent tools and frameworks, building practical familiarity with how they work. Progress to understanding the underlying principles and architectures. Eventually, move toward contributing to the global ecosystem, whether through developing locally-relevant skills, contributing improvements to open-source projects, or creating entirely new applications. This is not a race that requires starting from zero. It is an opportunity to build on foundations that the global open-source community has already laid.
A New Kind of Night Shift
Back in his apartment, Liu Shiyang has completed his configuration. His new AI system is already processing the three hours of interview recordings, and for the first time in weeks, he might actually get to bed before 2 AM. But as he watches the transcripts populate his screen, he begins to imagine something more ambitious. What if he could assemble not just a single assistant but an entire team of specialized agents? One agent could continuously monitor government announcements and social media for breaking developments on his beat. Another could specialize in analyzing financial disclosures and corporate filings. A third could help him identify patterns across years of archived news reports. A fourth could fact-check his drafts against multiple sources before publication.
This vision of a personal multi-agent system, a team of AI collaborators working together under the journalist’s direction, is no longer science fiction. It is the possibility that Skills.sh and Clowdbot have demonstrated and that the open protocol ecosystem has made achievable. The reporter who once spent countless late nights manually transcribing interviews and cross-checking facts can now imagine orchestrating a coordinated investigative operation.
What is happening in Liu Shiyang’s small apartment is a microcosm of a much larger transformation. Around the world, from technology companies to academic institutions to individual practitioners, people are beginning to explore what becomes possible when anyone can assemble their own AI army. The tools are open. The protocols are standardized. The costs are falling. For the first time in the brief history of artificial intelligence, the ability to build sophisticated AI systems is escaping the exclusive control of well-resourced research labs and technology giants. The age of agents is arriving, and it is arriving for everyone.