Author: Frank, PANews
OpenClaw, which has gone viral in the global AI community, has developed a "Chinese characteristic" version.
On March 6, 2026, an unusual scene appeared in front of Tencent Tower in Nanshan District, Shenzhen: Tencent Cloud engineers set up booths at the company’s headquarters, offering free OpenClaw installation services to passing developers and AI enthusiasts.
This open-source AI agent framework, nicknamed "Crayfish" due to its icon resembling a crayfish, is gaining public attention through a novel form of grassroots promotion.
In fact, the widespread popularity of "shrimp" has made it one of the fastest-growing non-aggregated software projects in GitHub's history, with its star count surpassing 250,000 in just a few months, outpacing established open-source projects like Linux and React.
Meanwhile, Tencent Cloud, Alibaba Cloud, JD Cloud, VolcEngine, and Baidu Intelligent Cloud have all launched one-click deployment services, while an industry known as "OpenClaw Installation Service" has quietly emerged, with remote installation fees ranging from 100 to 500 yuan; some claim to have earned 260,000 yuan in just a few days through this service.
When a free, open-source tool requires door-to-door promotion to gain traction, and when a product claiming "everyone can own an AI assistant" gives rise to a thriving business of paid installation services costing hundreds of yuan—what lies behind this frenzy: the opening act of the AI Agent era, or just another inevitable bandwagon trend destined to fade away?
If we rewind the clock twenty years, the story of a product called Little Smart might offer some insights.
OpenClaw is indeed a great tool, but it's not "JARVIS".
Before discussing OpenClaw’s fate, it’s important to recognize one fact: it is indeed an advanced product.
As an open-source AI agent framework, OpenClaw has accomplished what was once possible only for a select few tech elites: connecting the capabilities of large language models (such as Claude, GPT-4, and DeepSeek) to everyday tools like WeChat, Telegram, DingTalk, and Feishu through a unified interface. It is not merely a simple chatbot, but a “digital employee” capable of browsing the web, executing system commands, managing files, and writing code. As of March 2026, OpenClaw achieves 1.5 million weekly downloads on npm, and its plugin marketplace, ClawHub, hosts over 5,700 community-built skill packages with more than 1,000 active contributors.
This data sufficiently demonstrates that OpenClaw has truly addressed a key market pain point. Just as the PHS system emerged in 1998 and enabled ordinary wage earners to use a "wireless phone" for the first time, OpenClaw has given countless non-developers their first "AI assistant that can get things done." This value in market education should not be overlooked.
However, from the perspective of an ordinary user, OpenClaw is still far from the AI assistant J.A.R.V.I.S. envisioned in Marvel.
First, there is the barrier to installation and use. Deploying OpenClaw requires a Node.js environment, command-line operations, and API key configuration—conditions that present an almost insurmountable obstacle for users without technical expertise. This is precisely why the third-party installation industry exists.
More concerning are the hidden costs: users have reported that the installation and setup process alone consumed over $250 in API fees without yielding any useful results. Even after successful deployment, monthly token costs for heavy usage can range from $100 to $1,500—behind the words “free and open-source” lies a substantial hidden compute bill. For those without prior AI experience, it’s easy to turn a simple tool into a money-consuming monster. As a result, cost-saving guides have even emerged online, teaching users how to reduce token usage.
Secondly, security and stability. Since 2026, OpenClaw has been disclosed with multiple critical vulnerabilities: CVE-2026-25253 allows remote code execution via malicious links, CVE-2026-25157 involves operating system command injection, and the "ClawJacked" flaw permits malicious websites to hijack local AI agents via WebSocket.
Because OpenClaw requires extremely high system permissions (reading and writing files, executing shell commands, controlling browsers, and screen capturing), a security breach could have catastrophic consequences. A widely reported case involved Meta’s head of security, who, due to imprecise instructions, accidentally had hundreds of work emails deleted by an AI agent. China’s Ministry of Industry and Information Technology has also issued a security alert, urging users to guard against the potential risks of OpenClaw.
In addition, OpenClaw performs far less smoothly when handling complex tasks than shown in the demo videos. Multi-layer nested tasks can cause the large model to enter an infinite loop, and dense API calls often trigger rate-limiting mechanisms, leading to task interruptions. One user who attempted to automate their daily office workflows with OpenClaw summarized their experience: “Installed OpenClaw, spent the whole night troubleshooting. Burned through all my API credits—didn’t finish a single task.”
This feeling of déjà vu is just like the popular saying about Xiaolingtong phones twenty years ago: “Holding a Xiaolingtong, standing in the rain; switching it from left hand to right, but still can’t get through.”
From a product maturity perspective, today’s OpenClaw is more like an “AI that needs to be cared for” than an “AI that cares for you.”
As a developer with over two years of Vibe coding experience, the PANews author recently attempted to deploy a “crayfish” bot, but the experience was extremely underwhelming—it took half a day just to install Skills and connect Channels, and its capabilities were limited to basic tasks like checking the weather or setting calendar reminders. For more advanced programming, tools like Cursor or Antigravity offer greater control, directness, and stability. Furthermore, the automated execution hyped on social media can easily be achieved through API integration with large models and custom programs, not to mention better cost efficiency and control.
Who is driving this frenzy?
If OpenClaw's product strength deserves only a "mediocre" rating, why has it achieved such phenomenon-level popularity?
The answer may not lie in the product itself, but in the economics behind this frenzy.
The biggest beneficiaries are large model companies. OpenClaw is essentially a "token burner"—every task execution involves intensive calls to large language model APIs. The token consumption of a single OpenClaw agent far exceeds that of traditional conversational AI chatbots, making it a godsend for large model companies desperate for "usage growth stories." China’s large models and cloud services, thanks to their high cost-performance ratio, have gained international recognition and directly achieved token export.
API packages from some large model providers were once sold out, not due to insufficient supply, but because OpenClaw created an unprecedented density of demand.
Following closely are cloud service providers. OpenClaw emphasizes "on-premises deployment" to protect privacy, but for most ordinary users, purchasing a cloud server to run OpenClaw is a more practical choice. Tencent Cloud, Alibaba Cloud, JD Cloud, Volcano Engine, and Baidu Intelligent Cloud—all major Chinese cloud providers—have launched one-click deployment services for OpenClaw immediately. Alibaba Cloud has even introduced a dedicated "Coding Plan AI Coding Package" for OpenClaw users, offering fixed monthly fees to meet the surge in API demand driven by OpenClaw.
On Tencent Cloud's LightHouse Application Server, the number of OpenClaw users has exceeded 100,000. Tencent offers free installation in front of its headquarters, appearing as a charitable act, but in reality, it is a precise user acquisition strategy: you get OpenClaw installed for free, but you must pay continuously to run it on Tencent Cloud's servers.
This logic is identical to the low-price strategy used by telecom providers during the era of little smart phones: attract users with low entry barriers and retain them through ongoing service fees.
Another easily overlooked underlying factor is hardware requirements. OpenClaw’s recommendation for local deployment has directly driven demand for computing hardware. Overseas setup platforms like SetupClaw charge between $3,000 and $6,000, often including curated recommendations for specific hardware configurations. The operational logic of this supply chain bears a striking structural resemblance to the story of cordless phones (PHS) two decades ago, which spurred base station construction and ignited the entire telecommunications equipment industry.
Looking back at the history of Xiaolingtong, its rapid popularity in the Chinese market was not due to its strong product capabilities, but because China Telecom lacked a mobile communications license at the time and urgently needed Xiaolingtong, a "quasi-mobile" service, to expand its revenue streams. The driving force came from corporate business interests, not from consumers' inherent needs.
Today, OpenClaw is no different: large model companies need increased usage, cloud providers need server sales, and hardware manufacturers need shipments of computing devices. When a product’s popularity is driven more by supply-side forces than by actual demand, its prosperity is often fragile.
The ultimate form of AI automation: integration, not assembly
If OpenClaw is merely a transitional product for a phase, what should the true AI Agent look like?
The answer is emerging. 2026 is widely regarded by the industry as the "Year of the AI-Native Smartphone," as major tech giants are integrating AI Agent capabilities directly into operating systems and hardware, rather than requiring users to install third-party frameworks.
ByteDance, in collaboration with vivo and other smartphone manufacturers, has deeply integrated AI Agent capabilities into the underlying mobile operating system through the "Doubao Phone Assistant." Users can simply press the side button to enable the AI to autonomously perform complex tasks across apps, such as comparing prices and placing orders across multiple platforms, automatically booking meals or rides, and consolidating travel itineraries. The entire process runs automatically in the background, requiring no framework installation or API configuration.
On March 7, Xiaomi announced that Xiaomi Miclaw, built on its proprietary MiMo large model, has begun closed testing, with the goal of deeply integrating into the phone’s underlying system, accessing over 50 system tools, and ultimately controlling over one billion Mi Home smart devices. Overseas, Windows Copilot, Apple Intelligence, and Gemini in Android are pursuing the same path.
IDC forecasts that in 2026, shipments of next-generation AI phones in China will reach 147 million units, accounting for more than half for the first time at 53%.
This means that AI agents are evolving from a geeky toy that users need to assemble themselves into a ready-to-use, system-level capability.
Comparing OpenClaw with these native AI products highlights the differences clearly: OpenClaw requires users to set up their own framework, configure large model APIs, and connect individually to each platform—it is essentially a “universal adapter.” In contrast, native AI agents within smartphones and operating systems are ready to use out of the box, requiring no installation or configuration, with security fully guaranteed by the system manufacturer.
This comparison closely mirrors the relationship between PHS and 3G phones. PHS was phased out not because people stopped needing to make calls, but because 3G phones delivered calling functionality better, more portably, and with broader coverage. Similarly, OpenClaw may eventually be marginalized not because people no longer need AI agents, but because natively integrated AI agents will deliver an experience that OpenClaw can never match.
Echoes of the Past: Seeing OpenClaw’s Fate Through the Lens of Little Smart
Here, it is worthwhile to briefly revisit the lifecycle of Little Smart, which makes it clearer why OpenClaw is referred to as the Little Smart of the AI era.
The Xiaolingtong technology originated in Japan and was introduced to China by UTStarcom in 1998. At its core, it was not a mobile communication technology, but rather a wireless extension of landline telephony, using microcell base stations to connect user terminals wirelessly to the local fixed-line network. Its rapid popularity stemmed from one key factor: affordability. During an era when mobile phone calls were expensive and charged in both directions, Xiaolingtong’s one-way billing (free incoming calls) and low monthly fees enabled a large number of wage earners to use a “wireless phone” for the first time—earning it the nickname “the poor person’s mobile phone.”
In October 2006, the number of Little Smart users in mainland China reached a historical peak of 93.41 million.
However, technical flaws persisted. Poor signal coverage, lack of nationwide roaming support, and disconnections at speeds exceeding 40 kilometers per hour—“holding a Little Smart phone, standing in the rain”—was not a joke, but a real user experience. More critically, as mobile phone rates continued to drop and 3G technology matured, Little Smart’s only price advantage gradually disappeared. In 2009, the Ministry of Industry and Information Technology mandated that Little Smart networks be shut down by the end of 2011. By 2014, Little Smart base stations across mainland China had been progressively decommissioned, bringing an end to its 16-year journey.
Projecting the story of Xiaolingtong onto OpenClaw, three lines of reasoning are worth contemplating.
First, Xiaolingtong became popular not because it was superior, but because there were no better alternatives at the time. During the window when 3G phones had not yet been widely adopted and mobile service fees were high, Xiaolingtong offered a “sufficient and affordable” alternative. Today, OpenClaw faces a remarkably similar market environment: native AI agents are still immature, official agent products from large model providers are still under development, and OS-level AI integration has just begun. In this vacuum, OpenClaw fills the gap with a “free, open-source, and customizable” approach. But filling a gap is not the same as defining the future.
Second, the decline of PHS was not because it became worse, but because better technology emerged. PHS did attempt to evolve—it introduced MMS versions and tried to expand its coverage. Yet these improvements ultimately could not bridge the fundamental gap between its underlying architecture and true mobile communication. Similarly, OpenClaw can continue to iterate, add more skill sets, and optimize its deployment processes, but its fundamental nature as an “intermediate-layer framework” will not change. When DouBao Mobile Assistant enables users to perform cross-app operations with a single tap, when Xiaomi’s MiClaw can directly control all smart devices in the home, and when Apple Intelligence becomes a built-in feature of the iPhone, a third-party Agent framework that requires users to install, configure, and maintain it themselves—like PHS in the 3G era—is not worse; the world has simply changed.
Third, when China Telecom launched Little Smart, it wasn’t because it represented the future, but because it generated immediate revenue. Lacking a mobile communications license, Little Smart served as a workaround to enter the market. Today, cloud providers are investing in OpenClaw for the same reason: not because OpenClaw represents the future of AI, but because it can sell cloud servers, drive token consumption, and acquire users today. When better AI agent products emerge, these companies will switch focus with the same decisiveness as China Telecom did when transitioning to 3G.
However, any analogy has its limitations. It took 16 years for the little smart phone to phase out, but OpenClaw’s story has only just begun. The pace of AI technology iteration far exceeds that of telecommunications generational shifts, meaning the window for OpenClaw to transition from “hype” to “replacement” may be much shorter than that of the little smart phone—yet this also means the value it created for the industry during this window should not be entirely dismissed. It enabled hundreds of thousands of non-technical users to experience the potential of AI agents for the first time; its open-source ecosystem provided the community with a low-cost experimental platform; and the security, cost, and stability issues it exposed have offered invaluable lessons for future developers.
But history does not change direction because of popularity. At its peak, Xiaolingtong had 93.41 million users, yet its scale could not withstand the tide of technological advancement. OpenClaw has 250,000 GitHub stars, but star count has never been a measure of a product’s vitality. When AI capabilities are truly integrated into the phones, computers, and operating systems we use every day, and when an “AI assistant” is no longer a separate app to install but an infrastructure as ubiquitous as Wi-Fi, few will miss the “crayfish” that once required an entire night to set up.
In this nationwide craze for installing小龙虾, what truly deserves reflection is not what OpenClaw can do today, but whether we will be ready to embrace the true AI-native era when it is no longer needed.
After all, the little smart phone taught us a simple lesson: in the long race of technology, the winner is always the product that doesn’t require you to strain to adapt to it.