Grok Bot: AI Teammates With Their Own Computer
On 11 August 2026 SpaceXAI opened Grok Bot, in early beta, at x.ai/bot. The claim is specific: these are teammates you can give real work to. Each Bot has its own computer in the cloud, signs into the tools you already use, keeps working when your laptop is closed, and comes back when something needs approval.
That is a different product from a chat completion, and a different product from an n8n canvas. It is closer to hiring a junior who already knows the apps, then deciding how much unsupervised work you will tolerate.
The model underneath the launch is Grok 4.6, shipped a day later. Token prices for that model are a separate bill. Grok Bot's price is a seat.
Who Can Use It, And What It Costs
Today it is limited to SuperGrok Heavy, Cursor Ultra, and Cursor Teams Premium, on desktop and iOS. Enterprise is a waitlist.
Published seat prices on the product page:
| Plan | Price | What you get |
| --- | --- | --- |
| Cursor Ultra | $200 / month | Own computer, tool sign-in, scheduled routines, desktop and mobile, extended token limits |
| Cursor Premium Teams | $120 / seat / month | Ultra, plus team billing, a marketplace for skills and plugins, shared usage analytics, SSO |
If you are already on Cursor Ultra or SuperGrok Heavy, Grok Bot is included. If you are not, $200 a month is the entry ticket for one person. That is a lot of DeepSeek V4 Flash tokens. Do the arithmetic against the hours you actually expect to hand off before you buy a seat to "try agents."
What Is Actually New
Most agent demos still fail the last ten percent: they draft the email and leave it in a markdown file, they almost update the CRM, they almost file the ticket. Grok Bot's pitch is that the work lands in the real tool, because the Bot is logged in as you and driving the same UI.
SpaceXAI's own internal examples are the honest product spec: a sales Bot updating the CRM from call notes and drafting follow-ups, an ops Bot seating new hires and processing invoices from Gmail, an engineering Bot reproducing a bug in the product UI and handing the fix to another Bot. You can run several at once, put them in a group thread, and teach a routine by letting one watch you do the job once.
That last part is the operational risk as much as the feature. A Bot that learned your workflow by watching you will also learn your shortcuts, your exceptions, and your sloppy CRM hygiene. Show it the clean version.
What It Does Not Replace
Grok Bot does not give you an evaluation set. It does not give you an audit log you own. It does not make an undocumented process ready for automation. It does not decide what "correct" means for a Japanese client handover or a regulated inbox.
If the work is generic and already lives in SaaS tools (inbox triage, CRM hygiene, a first-pass outbound draft sitting in a review queue), a Bot is a reasonable trial, especially if the seat is already paid for. If the work is the product, or the data cannot leave the building, or you need to swap the model next month without rewriting the process, you still build a workflow you own. We wrote the teardown of that choice in AI agents vs AI workflows and when to move from n8n to custom software.
The same rule as every other launch this year: do not wait for the agent product to settle, and do not let the agent product become the process. Pin the job, measure the output, keep a human on the decisions that cost money.
If you want that measurement done on your actual stack before anyone buys a $200 seat, book a scoping call.
---
Sources: Introducing Grok Bot · Grok Bot product page · Introducing Grok 4.6