xAI has started renting out autonomous coworkers. The company introduced Grok Bot, a product that assigns each AI agent a persistent computer in the cloud, credentials for the software a business already runs, and permission to keep working after a person has closed the laptop. It is built for the kind of multi-step operational work companies currently route to junior staff, contractors, or nobody at all.
The pitch is specific: a bot signs into a CRM, an inbox, a ticketing system, even sites with no API or MCP connector, and returns only when it needs a decision a human has to make. That is a meaningfully different claim than a chatbot that drafts a reply and waits. If bots can close out real tickets and invoices unattended, it changes what a company staffs for.
xAI says Grok Bot began as an internal prototype that spread across the company before the public launch. According to the announcement, employees built bots to run outbound sales, campaign work in marketing, back-office operations, and bug triage, then started running several at once, with one bot coordinating specialist bots underneath it in a shared thread or group chat. The company quotes an internal product employee, identified only as Roman, describing the gap it is chasing: “There is a huge difference between 90% done and 100% done.”
Grok Bot enters beta today for SuperGrok Heavy subscribers and for Cursor Ultra and Cursor Teams Premium subscribers, on desktop and iOS. Enterprise customers can join a waitlist rather than get immediate access, a rollout pattern that suggests xAI wants broad usage data before it commits the product to accounts with stricter compliance demands.
This is a company announcement, not an independent evaluation. xAI supplies no benchmark, no completion rate, and no error rate for how often a bot finishes a job correctly versus how often a human has to step in and undo something. The testimonials in the post, credited to named employees in sales, community, growth, and operations roles, are marketing selections from inside the company that built the product (people with an obvious interest in the launch succeeding), not accounts from outside customers.
Strip away the branding and the shape is familiar. A persistent cloud machine per agent, with stored credentials and browser access to apps a person would normally use, is the same architecture OpenAI shipped with its computer-use agent and Anthropic shipped with Claude’s computer use capability, and it matches the sandboxed, always-on agent products now common across the market. Handing an agent a computer of its own is no longer the differentiator. What would separate one vendor from another is reliability across long, unsupervised runs and what happens when a credentialed agent acts on the wrong record: sends the wrong email, approves the wrong invoice, or edits a customer’s data incorrectly. xAI’s announcement does not describe rollback tools, audit logs, or who is accountable when that happens.
Teams evaluating Grok Bot should pilot it on reversible tasks first and ask specifically how failures get caught and undone, since the vendor has published only the successes.
xAI published this announcement, titled “Introducing Grok Bot,” on its own company blog; the post carries no publication date.
