Grok Bot: SpaceXAI launches AI agents with their own computer that sign in to your apps

SpaceXAI — the company formed by the xAI–SpaceX merger — launched the beta of Grok Bot on August 11, 2026, describing it as "AI teammates you can give real work to." This isn't another chatbot: it's the first joint product following the acquisition of Cursor, closed on August 14, and it aims squarely at the ground Anthropic opened with Claude Cowork.

The difference: every Bot has its own computer. A "Bot" is a named agent with its own memory, running on a persistent cloud virtual machine with a browser, terminal, and filesystem. That lets it sign in to your tools and use them the way you would — including the ones with no clean API or MCP connector — and keep working after you close the app or your laptop. The company's framing: there's a huge difference between 90% done and 100% done, and 100% happens when the work lands where a human would put it, inside the actual tool.

Many Bots at once, talking to each other. You can create one Bot per function — sales outbound, expense management, bug reproduction, chief of staff — and run them in parallel. They message each other, share context in threads or group chats, and hand off ownership of a task without you acting as the router pasting notes between conversations.

It learns by watching you. The "Teach a task" feature records up to 10 minutes of a workflow while you do it once. The Bot saves it as a skill (the method) and you can wrap it in a routine (the when): a fixed schedule — "weekdays at 8:00 AM" — or an event trigger like a Slack message or a GitHub notification. Each Bot supports up to 50 routines.

The fine print worth reading. Grok Bot authenticates through your Cursor account and requires cloud data storage: accounts on "Legacy Privacy Mode" must switch to a supported setting before it will start, and training opt-out follows whatever your Cursor account settings say.

There's another security detail that matters: all of your Bots share a single computer, scoped to your account rather than to each Bot. A login or a file you leave there is available to all of them, and deleting a Bot does not remove the shared files or browser sessions — you have to sign out and revoke access manually.

Where it does ask permission. By default, consequential actions require human approval: sending messages or invitations, publishing content, purchases and financial transfers, deleting or overwriting data, changing permissions, and production changes. Passwords, 2FA codes, and payment confirmations are never typed into chat: you take control of the agent's computer and enter them yourself.

Why it matters. 2026 is shaping up as the year agents left the chat window and moved into the tools. Claude Cowork did it on your desktop, with your local files; Grok Bot does it in the cloud, on a machine that doesn't shut down when you close your laptop. The underlying bet is the same — that you delegate work rather than request text — and the entry price says who it's for today: at $200 or $300 a month, this isn't competing with a $20 plan, it's competing with hiring someone.