George Zhang thought OpenClaw could have made him luxurious even though he didn’t really understand how AI agent virus software worked. However, he saw a video of a Chinese social media influencer showing how he could be used to manage stock portfolios and make investment decisions on his own. Zhang, who works in cross-border e-commerce in the Chinese city of Xiamen, was intrigued enough that he decided to try installing OpenClaw in overdue February.
Zhang is one of many people in China who have recently been caught up in the OpenClaw frenzy. Workshops teaching people how to employ the AI agent have popped up in cities across the country and attracted hundreds of people. Technology companies are racing to integrate OpenClaw with their platforms, and local governments have announced subsidies for entrepreneurs building products using it. Behind schedule last week, photos of grandpas and grandmas lining up to install the software went viral on the Internet.
After renting a cloud server from Tencent and purchasing a subscription to China’s multilingual model Kimi, Zhang was able to start talking to his OpenClaw agent, or his “lobster,” as many Chinese call theirs. Zhang tells me that he was initially impressed with the AI agent, watching how quickly it generated long-form market analysis based on breaking news. However, after a few days, his lobster started to sluggish down and instead of a detailed report, it only produced a basic outline of market trends. He asked OpenClaw to generate something similar to what it did on day one, to which the agent continually replied that he was “working on it” before never returning any results.
Zhang concluded that OpenClaw was not intended for people like him who had no coding skills. “It would tell me that I needed to configure the API port. However, this is a technical task and not something I can do unless I get a tutorial that walks me through it step by step,” he says. He eventually gave up trading Lobster stocks and instead asked him to aggregate AI industry news, which he used to build a social media content farm on WeChat.
This week, I spoke to six OpenClaw users in China about their experiences with the agent, and a clear picture emerged of the divide between adopters who are tech-savvy and those who aren’t. Those proficient in AI see OpenClaw as a productivity game changer, but non-technical people feel they were promised a wonderfully powerful AI product that ultimately fell low of expectations. But before the bubble burst, they had already started paying for cloud servers and LLM tokens.
The real factor driving OpenClaw mania in China is not ordinary users, but rather Chinese companies that stand to benefit financially from its widespread adoption. Major tech companies like Tencent, Alibaba, ByteDance, Minimax, Moonshot, and Z.ai have all seen AI productivity-boosting FOMO as a sporadic opportunity to get normal people to pay for AI services, and they are reaping the biggest benefits from it.
“A chatbot only uses a few hundred tokens per conversation; a single active OpenClaw instance can use dozens or even hundreds of times that number of tokens per day,” says Poe Zhao, technology analyst and founder of the newsletter Hello, Chinese technician. Every recent OpenClaw user is someone who pays 24/7 for LLM API calls. “That’s why Tencent engineers set up tables outside the company’s headquarters to help people install the software for free,” he says.
“I couldn’t understand any of it.”
Song Zhuoqun, a college student in China, says she started encountering problems with OpenClaw as soon as she tried to install it. Song is a social media intern at an artificial intelligence startup, but he has no programming experience, so figuring out how to get OpenClaw to work proved hard. She asked Doubao, ByteDance’s popular AI chatbot, to generate a step-by-step tutorial for her, but it didn’t assist much.
“There were pages full of code that I didn’t understand. I would just ask the AI to generate an answer for me, and then I would paste it in, run it, and an error would appear, so I would try a new answer,” she says. Installation turned out to be the most frustrating part of trying OpenClaw for Song, and I didn’t feel like I learned anything from it.