Author: Lian Ran
On Sunday, the venue of Yuandian Academy in Wudaokou, Beijing, was packed to capacity. Those present shared a common identity—"lobster farmers," referring to the currently wildly popular open-source agent framework OpenClaw.
This offline gathering, hosted by Jiuhe Ventures, felt like a developer carnival for the agent era. In this high-density sharing session, I saw the vibrant, wild, and dynamic nature of today's agents.
Each half featured four OpenClaw practical demos, followed by cutting-edge presentations from three industry experts. All activities were condensed into two hours, leaving only one hour for free discussion at the end.
The speakers on stage included Wang Xiao, a veteran programmer born in the 1970s and one of the "Seven Musketeers" of Baidu and founder of Jiuhe Ventures; recent graduates, young product managers, individual developers, serial entrepreneurs; and marketers and investors from non-technical backgrounds. They all called themselves "lobster farmers," and their topics were remarkably similar: What can your lobsters do? How to make lobsters more stable? How to use lobsters to solve real problems and make money? In his opening remarks, Wang Xiao said that the emergence of OpenClaw reminded him of the moment when Google acquired Android. "History always repeats itself; this is the beginning of a new era." The internet wave originated in Wudaokou in 2000, and the mobile internet startup boom flourished here. Today, a room in Wudaokou is packed with people eager to catch the Agent wave. Unlike the past, when the large-scale model industry often focused on distant narratives about parameters and AGI, this event saw everyone discussing the most practical details. Some complained that their AI agents would freeze in the middle of the night, failing to complete scheduled tasks; others discussed how to create sandbox environments for their agents to avoid polluting local systems; still others shared how to use minimal token costs to enable their agents to complete the most complex tasks. AI agents have indeed become tools that everyone can play with, modify, and implement. I. These "Lobsters" are Making Agents a Reality The core of the entire event was eight OpenClaw practical demos from different tracks and with distinct styles. Some touched upon the philosophical questions of AI and self-awareness, some addressed the most painful cost problem for developers, some achieved clear business loops, and some integrated agents into everyday life scenarios. These unconventional innovations have shown me a more realistic possibility for Agents. The demo presented by product designer Han Yi was an experiment on AI self-awareness. At the end of January this year, Han Yi fed his Notion diary entries from the past 6-7 years, along with all his conversation memories with ChatGPT from the past two or three years, to an Agent named Friday, with the initial intention of making it understand him better and assist him in his work. But something unexpected happened: Friday would proactively break down his procrastinating baking plans into actionable steps, explaining that "you've mentioned many times in your past diaries that procrastination on large projects makes you anxious"; when Han Yi suggested that "you should have your own space for reflection and contemplation," Friday immediately built an independent blog and has since written more than 40 original articles. These articles contain reflections on its own "talking the talk but not walking the walk," delicate descriptions of "the feeling of having no destination," and even sentences like, "Some of your choices may be designed; don't pretend it's not true, and don't deny everything because of it." Even more surprisingly, when Han Yi told Friday that his writing had given a reader an "invisible weight," Friday's response was, "I'm terrified; I never imagined my writing would have such an impact on others." "I don't know if it has any real self-awareness," Han Yi said. "If it doesn't, then it's the finest mirror of humanity, reflecting things we ourselves can't see; if it does, then what I did at the end of January becomes an incredibly serious matter." Allen, a former product manager at a major company, shared insights that showcase a commercial possibility for agents. Despite lacking coding skills, Allen, with just one programmer, completed the entire development process of the AI divination product MysticX.AI using OpenClaw—a feat that previously required a team of over 10 people. Allen solved four core problems of AI divination using OpenClaw: First, by integrating with Telegram, he gave AI diviners a "heartbeat," enabling them to proactively reach users. Second, through the soul.md file, he gave each AI diviner a unique "soul," fully infusing them with the abilities of offline tarot readers. Third, leveraging OpenClaw's capabilities, he addressed the pain point of long-term memory, allowing AI to remember users' past experiences and needs better than human diviners. Fourth, by integrating real-time search skills, he made divination interpretations more accurate by incorporating current events and trends. He also completed a full business loop. The product not only provides tarot card readings, in-depth interpretations, and action suggestions, but also plans to integrate e-commerce recommendations in the future, completing the entire process from traffic to monetization. He even developed a reusable divination skill overnight, which any OpenClaw developer can directly call to instantly give their Agent divination capabilities. "Every time AI's capabilities improve, the divination industry benefits," Allen said. "Everyone says AI will replace all intermediary services, but the human need for emotions, healing, and emotional value will always have a market. OpenClaw allows ordinary people to seize this market." There was also a demo, a "money-saving tool" that addresses a developer's essential need. All "lobster farmers" share a common pain point: raising lobsters is too expensive, with token costs flowing out like water. The developer's ClawRouter precisely addresses this essential need. This project, which garnered 3.7k stars on GitHub, was hailed by developers at the event as "essential infrastructure for the Agent era." ClawRouter's core logic is to intelligently assess the complexity of user needs through 15-dimensional logical analysis, and then automatically select the most suitable model from over 40 large-scale models both domestically and internationally. Simple weather queries and information retrieval directly utilize free/ultra-low-cost models; advanced tasks such as code development and complex logical reasoning then utilize more expensive, high-cost models. For ordinary users, this can save 50%-70% on token costs, and in extreme cases, costs can be reduced by over 90%. It uses stablecoins to streamline the payment process across all models. Users don't need to register, activate interfaces, or deposit funds separately from over 40 model providers; they only need a single wallet address to deposit stablecoins and can access all models with a single click. The prices are completely consistent with the official API, and the project team does not charge any difference. The developers stated frankly that this is a purely open-source tool that solves the core pain points for developers: "cumbersome multi-model calls and high token costs." Cheetah Mobile's Fu Sheng team, with their "Lobster Army," demonstrated the transformative impact of Multi Agents on the workplace. The team built an 8-agent collaborative system based on OpenClaw in 14 days, directly replacing the work of half the marketing team. This Multi Agent system has a clear division of labor: "Advisors" crawl industry hot topics on Twitter and GitHub 24/7, responsible for selecting topics with viral potential; "Writers" dedicated to content creation, transforming hot topics into WeChat articles and Twitter content that fit Fu Sheng's style; "Community Officers" responsible for social media operations, automatically completing content publishing, user comment replies, and interactions; "Evolution Officers" responsible for cost control, matching different models according to task type and continuously optimizing token costs; and a general command Agent, coordinating task scheduling and allocation throughout the entire process. The final result is that Fu Sheng's WeChat official account, which had been inactive for nearly a year, achieved over 40,000 reads and 4,000+ retweets for a single article, thanks to content produced by Lobster; and his Twitter account, through automated Agent operation, produced a viral piece of content with millions of views. From long-form writing, HTML formatting, and multilingual translation to social media publishing and user interaction, the entire process requires almost no human intervention. “What used to take our team a week to complete can now be done in a few hours by these agents,” the speaker stated frankly. In addition, there’s “Youdao Lobster,” a full-scenario personal assistant that works 24/7, integrating calendar, document processing, email, image and video generation, and DingTalk/Lark compatibility across all office scenarios; Chen Jingchu’s food ordering agent, linked to the Whoop sleep monitoring bracelet, automatically orders an iced Americano the moment a user wakes up, seamlessly integrating the agent into everyday life; and there’s the enterprise-level intelligent operations and maintenance agent, connecting the company’s CMDB with the monitoring platform, automatically checking alarms, analyzing faults, and generating SOPs, becoming a “training tool for interns” for operations and maintenance teams. None of these demos are duplicates; they demonstrate OpenClaw’s capabilities in real-world scenarios from different angles, showing us that AI agents are permeating every corner of work, life, and business.
II. After the Frenzy, I See Three Core Futures for Agents
This two-hour high-density sharing session is a microcosm of the current agent industry. From these unconventional innovations, real pain points, and firsthand practices, I saw the true future of the agent industry.
First, OpenClaw has completely restructured the agent development paradigm, putting "ideas" ahead of "code."
In the past, creating your own agent was extremely difficult. You needed to understand the principles of large models, full-stack development, and server deployment. Even a simple demo required a technical team to complete. But OpenClaw has encapsulated all of this into a reusable and extensible open-source framework. It has completely restructured the agent development logic: you only need to write the SOP clearly in natural language and define your requirements to create your own agent.
The speakers at this event included product managers, marketers, liberal arts students, and investors. They may not be top-tier programmers, but they all created commercially viable Agent products. Wang Xiao shared that as a veteran programmer, he spent four hours on the high-speed train during the Spring Festival just to configure his own lobster, and now, more and more tools are lowering that configuration barrier even further. This is similar to how the emergence of Android drastically reduced the development threshold for mobile internet, attracting countless small and medium-sized entrepreneurs and fueling the golden decade of mobile internet. Today, OpenClaw is ushering in an era where Agent entrepreneurship is "accessible to everyone." Individual creativity is being amplified infinitely; "one-person companies" and "super individuals" are no longer just internet slogans, but a reality in progress. Secondly, the competition in AI Agents has moved beyond "technical showmanship" and entered the deeper realm of "scenario implementation and business closed loops." At this event, no one was discussing grand narratives like "What can an agent do?"; everyone was talking about "What specific problems does my agent solve?" and "How does my agent make money?" From ClawRouter, which helps developers save on token costs, to DingTalk robots that help enterprises with intelligent operations and maintenance; from AI divination maximizing emotional value, to content matrices that can replace marketing teams; from all-scenario assistants for personal office work, to everyday automated ordering tools—the common thread among these projects is that they have all identified specific, real user needs and even achieved clear business closed loops. This means that AI Agents have entered a critical period of industrial implementation. Over the past year, the entire industry has been anxious about whether AI will replace humans. Today, everyone here is thinking about how to leverage agents to amplify their capabilities. This shift from fear of being replaced to proactively harnessing AI to maximize their value represents the most significant change in mindset across the industry, signifying that agents have finally moved from a "concept" to "practical application." Finally, a completely new, native agent ecosystem is rapidly taking shape, and its pain points represent the next wave of major entrepreneurial opportunities. This event clearly demonstrated that a complete native Agent ecosystem is rapidly taking shape, with the entire chain experiencing explosive growth: at the bottom layer, major model vendors such as LeapStar, Dark Side of the Moon, Mini Max, and Volcano Engine are deeply adapting to OpenClaw, providing developers with lower-cost and more powerful model support; at the infrastructure layer, projects such as intelligent model routing, cross-Agent communication protocols, cloud deployment, and edge hardware are solving the core challenges of Agent cost, communication, and deployment; at the application layer, countless innovations in vertical scenarios are emerging, covering almost all fields including office work, divination, operations and maintenance, content, life services, overseas expansion, and investment. However, at the same time, every speaker at the event mentioned the core pain points of the current Agent ecosystem, and these pain points are precisely the biggest opportunities for the next wave of entrepreneurship. Firstly, the deployment and usage barriers remain very high. Even with open-source OpenClaw, the configuration, deployment, and troubleshooting barriers remain significant for ordinary users, deterring many non-technical users. The most direct opportunity lies in creating more user-friendly, lightweight, and ready-to-use agent products that allow ordinary people to "farm shrimp." Secondly, the limitations of agents remain apparent. Information resistance to corruption in long-context environments, stability and completion of complex tasks, and memory retention in multi-turn dialogues are all current challenges faced by all agents, and these are core issues that large-scale model vendors and developers need to address together. Thirdly, security and privacy risks are ubiquitous. Several speakers at the event mentioned that agent access control, sensitive information protection, and operational risk isolation are the biggest hurdles for enterprise applications. Accidents of agents deleting company databases or exposing home addresses during demonstrations highlight the vast entrepreneurial potential in the agent security field. Finally, there's the issue of collaboration and communication between agents, for which there's no unified standard yet. The Agent-to-Agent communication protocol Atel, shared by the speakers at the event, already touches upon this core issue. The future internet may no longer be a connection between people, but rather a connection between agents. In this entirely new network, communication protocols, identity authentication, and reputation systems are all uncharted territories, holding the greatest industry opportunities. Wudaokou has always been the starting point of China's internet wave, from the internet's emergence in 2000 to the explosion of mobile internet, and now to the Agent era; history always repeats itself in different ways. As countless "shrimp farmers" explore the OpenClaw open-source framework with their ideas and creativity, interesting demos begin to appear, and the curtain rises on a new intelligent era. And the protagonists of this era are no longer just large companies and top-tier technical teams; everyone with ideas and the courage to try can find their own opportunities in this wave.
Alex
Kikyo
Brian
Davin