A grassroots craze for an open-source AI agent nicknamed "lobster" has rippled from developer forums into the offices of Shenzhen’s municipal government and the sidewalks outside Tencent’s headquarters. OpenClaw, an agent that users say can "take over a computer and free your hands," sparked long queues after cloud engineers offered free one‑stop installation help. Entrepreneurs quickly monetised the demand: paid installers now advertise on platforms for on‑site and remote setup, with prices commonly between 300 and 1,000 yuan and some claiming substantial short‑term earnings.
The phenomenon attracted comment at China’s national legislature. Gao Wen, an academician and head of Pengcheng Laboratory, told a Guangdong delegation meeting that even Pony Ma, Tencent’s founder, had been surprised by how quickly the tool caught on. Gao urged Guangdong—particularly Shenzhen—to seize the moment, arguing that early technical deployment could draw global participation and strengthen the region’s industrial influence.
Local governments have already moved beyond pilot chatter to deployment. Futian district officials said they have put OpenClaw‑driven assistants into two civil service roles: a "public sentiment analyst" that ingests and summarises citizen complaints, and a "sanitation permit counsellor" that chats with applicants and pre‑checks paperwork. Officials reported efficiency gains of two to three times for routine processes and emphasised the system’s intellectual‑property provenance: the agent was developed by a local company and claims full proprietary rights over its district deployment.
Tencent has sought to channel the craze into its ecosystem. The company updated QQ’s developer platform to make it trivial to create QQ bots that communicate with OpenClaw, allowing ordinary users to bind chatbots to agent instances and exchange multimedia messages through familiar interfaces. That integration lowers the technical barrier for everyday adoption, turning hobbyist interest into mainstream utility.
The surge highlights both opportunity and risk. OpenClaw is reportedly more demanding to run than zero‑prompt models such as ChatGPT; it uses token feeding as part of its learning and operation model, which users metaphorically describe as "raising" the lobster. That concept, combined with local installers and rapid deployment in public services, raises questions about data governance, model auditing and security. Regulators and commentators have already flagged potential safety and privacy risks surrounding mass, decentralised use of agents that can operate with system‑level permissions.
Beyond technophilia and entrepreneurial hustle, the episode is a test of how cities and companies govern emerging AI tools. Shenzhen’s pragmatic embrace—pairing municipal pilots with local development and a major platform’s integration—could accelerate a model of civic AI deployment that other Chinese cities watch closely. At the same time, the patchwork of paid instalment services, volunteer clinics and rapid adoption could exacerbate uneven standards and create avenues for misuse if oversight does not keep pace.
For an international audience, the story is a useful signal of how quickly open‑source agent software can permeate public administration and consumer markets when platform players and local government align. The speed of adoption underscores China’s capacity to mobilise private‑public ecosystems around nascent AI capabilities, while also illuminating the governance dilemmas that will shape global debates about responsible agent deployment.