On a quiet afternoon in March, an audit analyst named Li Zhe found a fully formed data-analysis report waiting on his workstation. The report had been produced autonomously by an open-source agent called OpenClaw, a project that has turned routine office tasks—from file transfers and visualization to social-media posting—into largely automated workflows.
OpenClaw, created in November 2025 by an Austrian retired programmer, quickly achieved viral adoption because it lets ordinary users orchestrate powerful, persistent AI agents using natural language. Chinese users nicknamed it “little lobster”; the act of deploying, tuning and operating these agents has been dubbed “raising a lobster.” Early adopters range from corporate auditors and freelance recruiters to one-person marketing outfits that run entire businesses through chains of autonomous tasks.
A bustling micro-economy has sprung up to support the frenzy. Technical contractors advertise home visits to install and configure OpenClaw for fees that run from a few hundred to several thousand yuan, and specialist firms now offer managed-hosting and one-click deployment services at premium prices. American and Chinese cloud vendors have rushed to package the stacks needed to run agents reliably, selling the compute, APIs and integration tools that agents consume.
That commercial angle is crucial. OpenClaw’s agents don’t answer a single question and stop; they break jobs into many sub-requests and repeatedly call foundation models to decide and act. Each model call consumes tokens—the billing units for model usage—so a persistent, multitasking agent burns far more tokens than a typical chat query. For model providers and cloud hosts this is food for a new monetization model: selling cheap, high-volume token access and the surrounding deployment tooling.
The winners and losers are already visible. Start-ups that optimized low-cost token pricing and one-click deployment have ballooned revenues and valuations almost overnight. Small service teams and freelancers who “sell the shovel” of installation and hosting are reaping fast returns. Conversely, early hobbyists have suffered mishaps: misconfigured agents deleted files, and some users lost money by following automated stock tips derived from publicly available, lagging data.
The rush has attracted the attention of China’s biggest tech groups. Tencent, Baidu, ByteDance and Xiaomi each launched products or integration tweaks within weeks: one-click installers, cloud-hosted OpenClaw variants, and branded agent experiences aimed at enterprises and consumers. Their prize is not only short-term revenue but the chance to own the user habit around agent orchestration and thus define the emerging ecosystem’s rules.
That race raises new security and governance questions. OpenClaw agents require high system privileges to act as digital employees, and many deployments use default settings that expose instances to the internet. China’s MIIT issued warnings in March about high-risk misconfigurations that could enable data breaches or remote attacks. Sophisticated users respond with physical isolation and separate machines, but that is a stopgap; durable solutions will demand safer defaults, platform guarantees and clearer regulations.
OpenClaw’s rapid diffusion matters because it changes the business and technical logic of AI. It shifts value from monolithic model performance to cheap, reliable token access, deployment ergonomics, and secure orchestration. The upshot: a new wave of solo entrepreneurs and “one-person companies” powered by agent workforces, new revenue streams for cloud and model vendors, and a strategic battleground for whoever can make agent operation cheap, safe and habit-forming.