Tencent’s chairman Ma Huateng unexpectedly turned a niche open‑source project into headline news when he shared the company’s new product matrix for OpenClaw — a family of locally runnable AI agents that Chinese netizens have nicknamed “cyber lobsters.” The post underscored Tencent’s push to offer a spectrum of agent options — from locally hosted variants to cloud and enterprise editions — and helped accelerate a wave of consumer curiosity and experimentation.
OpenClaw is notable because, unlike the cloud‑hosted chatbots familiar to most users, it can be deployed on an ordinary PC and instructed to carry out concrete tasks such as sorting files, handling email or writing code. That hands‑on capability has created a small service economy: local tech vendors offering to install agents for customers, charging between roughly RMB 300 and 800 for in‑person installs, or RMB 50–200 for remote setup.
The boom has produced a countertrend. Users report problems ranging from accidental deletion of files to privacy leaks and unauthorized spending when agents misbehave. China’s National Internet Emergency Response Center issued a security advisory urging caution when deploying OpenClaw, and social media has been filled with requests for help fully removing the software. Entrepreneurs quickly monetized that anxiety with paid “uninstall” services priced far lower than installation, prompting jokes about a self‑feeding “cyber loop” of pay to install, pay to uninstall — and reminding observers that many people who never joined the craze effectively “earned” the installation fee by sitting out the fad.
The OpenClaw episode matters beyond a local meme. It illustrates a broader industry shift that accelerated in 2025: the race from generic large language models to vertically specialised, proactive agents that act on users’ behalf. Firms such as Tencent expect these agents to evolve from passive question‑answer systems into autonomous helpers that can perform everyday digital labour. That transition has produced a rapid, iterative burst of projects in China and globally, from agent toolkits to specialised content generators.
Yet the security and utility calculus is mixed. Local deployment exposes users to misconfiguration and possible backdoors, especially when setup requires technical know‑how; those risks explain the quick emergence of paid installation and removal services. At the same time, major platforms are likely to offer cloud‑hosted agent services that reduce deployment risk but increase centralisation and attendant privacy and vendor‑lock‑in concerns. For most everyday consumers the practical gains remain modest; many will treat agents as a novelty similar to short‑lived app fads rather than indispensable tools.
Looking ahead, the competition over “who finds the right entry point” will intensify. The immediate winners will include security firms, cloud providers and system integrators who can package agents safely and prove real productivity gains. Regulators and consumer advocates will play a growing role if incidents multiply, while public perception will determine whether autonomous agents become routine workplace tools or a passing curiosity amplified by AI anxiety and social signalling.