SpaceXAI launched Grok Bot in early beta on 2026-08-11, describing it as “AI teammates you can give real work to.” The product ships as a desktop and iOS app and is available to SuperGrok Plus and Heavy subscribers plus Cursor Pro+, Ultra, and Cursor Teams Standard and Premium tiers. Enterprise customers are on a waitlist. The company says Grok Bot began as an internal prototype that spread across the company before being opened up.
The design choice that distinguishes it is stated directly: Bots “share a computer of their own in the cloud,” so work does not stall when the user steps away, and they sign into and operate the apps and websites a person would - explicitly including “platforms with no clean API or MCP.” That is a bet that browser-and-desktop operation, not tool integrations, is the general path to getting work into the systems where it actually has to land. SpaceXAI’s internal examples are mundane and specific: a sales Bot writing call-transcript notes into a CRM and drafting follow-ups, an ops Bot seating new hires and processing invoices arriving in Gmail, an engineering Bot reproducing a UI bug, filing the ticket and handing the fix to a second Bot.
The second design choice is multiplicity. SpaceXAI says its own staff commonly run several Bots in parallel with one acting as a chief of staff over specialists for inbox, expenses, recruiting or bug fixes. Bots can message each other, share context in threads, and sit in group chats where they pass work and assign ownership among themselves, pulling the human in only for judgment calls. There is no workflow builder: the interaction model is messaging a Bot the way you would text a colleague, and picking the thread up later on another device.
For a business leader the claim worth testing is the one SpaceXAI puts in a pull quote from its own product lead - that most AI gets to 90 percent and Grok Bot “can finish the swing, because the work lands where a human would put it, in the actual tool.” That is precisely the part an early beta cannot yet evidence, and agents that hold live credentials to a CRM, an inbox and an invoicing system carry an audit and blast-radius problem that no benchmark measures. The interesting signal is directional: a frontier lab is now selling seats for agents rather than tokens for a model, and pricing them alongside human-style subscriptions.