First Wave of "Lobster" Farmers Are Already Thinking of Paying Someone to Unload

Viral fever, quick adoption — and fast buyer's remorse

OpenClaw (“龙虾”), an open‑source AI agent framework developed by Austrian programmer Peter Steinberger in 2025, went from niche developer talk to national obsession in a matter of weeks. It can be set loose to manage email, scrape and summarize information, post to social feeds and run background workflows — a 24/7 digital assistant that promised to be the perfect, ever‑compliant employee. Reportedly people queued for free installs outside tech campuses in Shenzhen, and some claimed to have made heavy money deploying or installing instances for others. But enthusiasm curdled almost as fast as interest spiked. Who thought we’d soon see people paying others not to put these agents on their machines?

Costs, complexity and the new gig economy

The core appeal was obvious: automation that feels like hiring an extra worker without payroll. The reality is messier. OpenClaw itself is free, but it calls cloud AI models and consumes paid "tokens" — users say bills can climb quickly. One experienced developer told Huxiu his company spends roughly $100 a week for a shared instance; other online anecdotes (it has been reported that) show single nights burning millions of tokens and invoices in the low‑thousands of dollars. Deployment is nontrivial too: proxy, language, permission and integration issues mean many ordinary users pay 100–300 yuan on marketplaces like Taobao (淘宝) and Xianyu (闲鱼) for professional installation — and, increasingly, for removal.

Security, regulation and a sharply divided verdict

Security and platform policy have shaped the backlash. It has been reported that OpenClaw plugins can be weaponized to steal keys or change network settings; users reported devices going offline after the agent altered configurations. Google moved to block OpenClaw from calling its large models in late February, and on March 10 China’s National Internet Emergency Response Center issued a safety advisory citing prompt‑injection, inadvertent deletions and other high‑risk behaviors. Social platforms such as Xiaohongshu (小红书) quickly banned AI‑managed accounts, undercutting hopes that automation could be a shortcut to social fame. For some companies and power users OpenClaw remains cost‑effective; for many first adopters it has been a costly, technical and security headache. Is this another ephemeral "year of" tech craze, or the painful growing pains of a new class of tools? The answer is still unfolding.