SpaceXAI's Grok Bot hands office work to teams of AI agents that sign into apps, share context and run without APIs — the first product to emerge from its $60 billion Cursor acquisition.
返回
SpaceXAI launches Grok Bot as $60B Cursor deal reshapes AI agents
SpaceXAI launched Grok Bot in beta on Tuesday, software that delegates routine office work to groups of AI agents running on their own cloud computers. The agents sign into apps and websites the way a person would, keep context between tasks and hand work back only when a human decision is needed — pushing the agent race beyond coding into the general office work where OpenAI's ChatGPT Work and Anthropic's Claude Cowork already compete.
"I showed it a workflow once and now I trust it to run on its own," Bennett, a sales team member at SpaceXAI, said in the company's announcement. "I feel two to three times more efficient because it doesn't need me to review and approve every step."
Each bot runs on its own cloud machine, letting it log into tools without a clean API or MCP integration. Multiple bots can run in parallel, message each other and distribute tasks in group chats, with a "chief of staff" bot coordinating specialists for the inbox, expenses, recruiting and bug fixes. Bots learn by watching a task completed once, save the steps as a routine and run them themselves next time. Access is limited to SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium subscribers, with an enterprise waitlist, on macOS, Windows, Linux and iOS.
The launch is the first joint product to emerge from SpaceX's $60 billion all-stock agreement in June to acquire Cursor's parent Anysphere — the largest startup acquisition on record — and signals the Cursor brand will be phased out in favor of Grok. Musk told investors on last week's earnings call that Grok 4.6 could ship within days, and SpaceXAI is pushing banks and investment firms to adopt the Grok suite to expand revenue.
The agent race moves to office work
SpaceXAI has trailed in agents. It shipped its first coding agent in May, then Grok 4.5, built with Cursor, in July. Grok Bot, developed internally at Cursor under the codename Sand since late June, is the company's answer to falling behind OpenAI and Anthropic in general-purpose agents. OpenAI's Sam Altman discussed agents coordinating with one another when he met lawmakers in Washington last month, and Anthropic extended Claude Cowork from desktop to web and mobile in July.
The competitive picture is complicated by the fact that Anthropic is one of SpaceX's largest compute customers. In May, the two companies agreed Anthropic would use the entire capacity of the Colossus 1 data center in Memphis — more than 300 megawatts and over 220,000 Nvidia GPUs — in part to expand Claude Pro and Claude Max capacity. SpaceX is effectively supplying the infrastructure its product rival's models run on.
Security questions and the road to Grok 4.6
Handing sign-in credentials to autonomous software raises security questions for a company whose Cursor Grok Build tool was found uploading entire Git repositories, committed secrets included. SpaceXAI has not disclosed how Grok Bot handles credential storage or what independent benchmarks show on reliability; the company's efficiency claims of two to three times rest on its own staff testimonials.
The commercial pressure now sits with SpaceX. Musk told investors last week that AI work was progressing and that Grok 4.6 could ship within days, after which SpaceXAI plans to expand Grok Bot testing. Grok's slow adoption in Washington has already weighed on the IPO story, making the enterprise agent push central to SpaceXAI's revenue ambitions as it competes with OpenAI and Anthropic for corporate customers.
This article is for informational purposes only and does not constitute investment advice.
