After racking up 270,000 GitHub stars in two months, OpenClaw quickly ran into security warnings, uninstalls and questions over high costs. Its retreat shows that whether AI assistants can move into the mainstream no longer depends on demo performance, but on cost, reliability and the barrier to adoption.
What this is
OpenClaw is an AI assistant focused on automating everyday tasks, such as organizing email. Its main selling point was becoming a “24/7 digital assistant,” but actual use often requires users to install an environment, obtain an API key—a credential developers use to call models and external services—configure browser permissions and run local services.
The hands-on tests cited in the original article show that monthly Token—the computational resource consumed when a model processes text—costs can exceed RMB 10,000, with some overseas users spending as much as USD 10,000. Another critical issue is the boundary of permissions: when an assistant can delete emails, make purchases or invoke tools, a single mistaken decision can cause real losses.
Industry view
We believe this cooling of interest is not bad for the industry. AI products are beginning to move from demo competitions centered on “Can it be done?” to operating tests of “Can it be done reliably, cheaply and safely?” Only when product design addresses installation, task success rates, running costs and permission controls can AI become business infrastructure.
But the risks are clear: OpenClaw is not an isolated case. Other assistants may face the same high model-inference costs, heavy dependencies on complex environments and runaway permissions. Some developers still argue that it offers value to people comfortable with command-line tools and professional workflows, so its retreat cannot simply be treated as a technical failure. The industry will ultimately leave two types of products: professional tools and lightweight services that genuinely lower the barrier to use.
Impact on regular people
For enterprise IT: When enterprises procure AI assistants, they need to assess permission approvals, logging and spending caps, rather than comparing model capabilities alone. Only tools that can integrate with existing systems and clearly explain how they handle errors are suitable for core workflows.
For individual professionals: People can still use OpenClaw for low-risk tasks such as organizing email and summarizing information, but they should not directly authorize payments, deletions or external sending. Whether the efficiency gains hold up should be measured by actual time spent and API costs.
For the consumer market: “Available to everyone” is becoming a product constraint rather than a marketing slogan. Consumers will pay closer attention to out-of-the-box usability, transparent subscription pricing and whether mistaken actions can be reversed. These factors will influence purchasing decisions more than conceptual demos.