xAI has launched Grok Bot in early beta, introducing a service it describes as a team of always-on AI agents that can work across applications, inboxes and websites from their own cloud computers. The release moves Grok beyond a single conversational model into a product aimed at delegating multi-step operational work while a user is away.
According to xAI, a Bot can sign into the tools a customer already uses, complete an assigned job end to end and return when human approval is needed. The company says people can message a Bot from desktop or mobile, continue the same conversation across devices and use multiple Bots in parallel. These are ambitious capabilities, so the beta status and the limits of initial availability are important parts of the announcement.
A separate computer for each agent
xAI says each Bot has its own cloud computer. That is designed to let a task continue when the user closes an app or steps away, and to allow the agent to work with software that does not present a clean API or Model Context Protocol integration. In the company’s framing, the result should feel less like a chat session and more like handing work to a colleague.
The distinction matters because browser and computer-use agents carry different operational risks from text-only assistants. A system that can log in, navigate software and take actions can be useful for repetitive work, but it can also encounter sensitive data, irreversible actions, access controls and ambiguous instructions. Organisations considering the beta will need to decide which accounts, permissions and approval points are appropriate before asking an agent to operate in production systems.
From a prompt to an ongoing hand-off
xAI says Grok Bot is designed to retain context about a user’s preferences and past work. The company describes a workflow in which a person shows a Bot how to complete a process, the Bot records that workflow as a routine, and subsequent tasks can be handed over without repeating every instruction. It also says Bots can follow up on stalled work and know when to request a decision.
That proposed memory and routine layer is central to the product’s appeal. It could reduce the setup required for recurring tasks, but it also raises practical governance questions: how routines are reviewed, how a user can inspect or revoke saved instructions, what happens when an underlying application changes, and how access is removed when a team member leaves. The announcement does not answer those implementation questions in detail, so they remain points for prospective business users to verify during the beta.
Multiple agents can share a piece of work
xAI also describes group coordination. It says several Bots can work at once, message one another, share context in threads and be placed in a group chat. The examples supplied by the company include a sales Bot updating customer records and drafting follow-ups, an operations Bot processing invoices and onboarding work, and engineering Bots that reproduce a product issue, file a ticket and pass it to another Bot for debugging.
Those examples illustrate why multi-agent systems are attracting attention: a task can be divided into specialist roles rather than routed through one assistant. They should still be treated as xAI examples, not a guarantee of performance. The more agents touch the same customer data or operational workflow, the more important it becomes to define ownership, verify outputs and keep a clear audit trail of decisions and actions.
Who can use it now
Grok Bot is available in beta to SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium subscribers on desktop and iOS, according to xAI. Enterprise customers can join a waitlist for future access. The company says the product began as an internal prototype, where teams used it for sales outreach, marketing, office operations and bug fixes, before the external beta.
The release is therefore not a general availability commitment for every Grok customer. Availability by plan, platform and region may change as the beta develops. Teams should confirm the current subscription requirements, supported applications, data-handling terms and controls directly with xAI before relying on the service for customer-facing or regulated work.
A consequential expansion of the Grok product line
Grok Bot is notable because it packages model capability with persistent computer access and a collaboration model intended for ongoing delegation. If it works reliably, that could make it a useful option for teams looking to automate bounded, repeatable workflows. The same features make careful rollout essential: begin with low-risk tasks, give the agent only the access it needs, require approval for consequential actions and review its behaviour before broadening its role.
For now, the launch is an early-beta product announcement rather than evidence of a mature autonomous-workforce platform. xAI has made a clear claim about where it wants Grok to compete; the useful next evidence will come from the controls, reliability data and customer results that accompany wider availability.
Practical boundaries for an early beta
Before assigning any agent a real task, a team should identify the systems it may open, the information it may read and the actions it may take. A Bot working in an inbox or CRM could encounter customer records, contracts, invoices or credentials. A cautious trial would use a limited account, a non-production environment where possible and clear rules for when the Bot must stop for approval. Teams should also test error recovery: whether the Bot records what it did, how a person can correct a routine, and how access is withdrawn promptly if a workflow proves unsuitable.
These controls are not a criticism unique to Grok Bot; they are standard requirements for any computer-use agent. They are especially relevant here because xAI’s product is intended to remain active while the user is away and to work across several tools. The beta may be useful for learning where agents reduce repetitive work, but it should not be confused with permission to remove accountable human oversight.