TL;DR. Grok Bot, launched on 11 August 2026, gives you persistent AI agents that have their own computer, sign in to your tools, and keep running when your laptop is closed. Three things in the official documentation should shape your setup before you create a single Bot. First, the cheapest door in is Cursor Pro at $20 a month, not SuperGrok at $30, which is $120 a year less for the same Grok Bot access. Second, Prompts, Skills and Routines are not three separate features: the design documentation defines a Prompt as something that can be used once, saved as a Skill, or triggered automatically as a Routine, which makes them one object at three stages of maturity. Third, and most consequentially, every Bot in your account shares one persistent cloud computer, and the documentation states plainly that isolation is per user, not per Bot. That single sentence should determine how you scope logins before you appoint a Chief of Staff.
What Grok Bot actually is
Grok Bot is SpaceXAI’s persistent agent product, announced on 11 August 2026 and described as “AI teammates you can give real work to.” The product page’s framing is that Bots “sign in to your tools, use them just like you do, and come back with finished work.”
The design documentation published on 3 September 2026 is more precise, and it is the document worth reading before you set anything up. It defines five objects:
- Bots are “persistent agents with their own identity, memory, runtime, and tools.”
- Chats are “the conversational interface for working with a Bot.”
- Prompts “give a Bot context or instructions. They can be used once, saved as Skills, or triggered automatically as Routines.”
- Tools “let Bots access information and take action through software, APIs, connectors, the shell, or computer use.”
- Artifacts are “documents, designs, code, data, and other durable outputs that Bots create or modify.”
The stated design principle behind all of it: “As agents take on more responsibility, the interface should ask less of the person.”
Grok Bot runs on desktop, on macOS or Windows, and on iOS.
Step 1: Pick the cheapest door, not the obvious one
Most coverage frames Grok Bot as a SuperGrok feature. The pricing pages say otherwise, and the difference is worth real money.
Table 1. Plans that include Grok Bot access (official pages, read 22 September 2026; annual cost calculated by CEOtudent)
| Plan | Published price | Annual cost | What the page lists |
|---|---|---|---|
| Cursor Pro | $20 / month | $240 | Grok Bot’s own computer, signs into your tools, Routines on a schedule, works on desktop and mobile, weekly Grok Bot usage included |
| SuperGrok | $30 / month | $360 | Grok Bot access, Grok 4.6 model, higher rate limits, connectors |
| SuperGrok Plus | $100 / month | $1,200 | Significantly higher usage, 1080p video, priority access |
| Cursor Teams Standard | $40 / seat / month | $480 / seat | Everything in Cursor Pro+, centralised team billing, team marketplace for skills and plugins, shared usage analytics |
Annual figures are the published monthly price multiplied by twelve. Grok Bot is also available to Cursor Pro+ and Ultra, SuperGrok Heavy, Cursor Teams Premium, and Enterprise, and the site states: “Already on an eligible Cursor, SuperGrok, or Teams plan? Grok Bot is included.”
Two decisions follow. If you want Grok Bot and nothing else, Cursor Pro at $240 a year is the cheapest entry point and saves $120 a year against SuperGrok, a third off the price. You would choose SuperGrok instead when you want the Grok 4.6 model, higher rate limits across chat and voice, and connectors rather than agent access alone.
One caveat that the pricing table does not show: subscriptions “come with weekly usage included, with additional usage billed based on token cost.” The subscription is a floor, not a cap. Budget for the agents actually doing work.
Step 2: Understand the one architectural fact that changes everything
Before you create Bots, read this FAQ answer from the product page in full:
“Do Bots share one computer? Yes. Every Grok Bot shares one persistent cloud computer. Your Bots share that machine (files, browser, logins), so they can hand work off and keep context. Isolation is per user, not per Grok Bot.”
This is the most important sentence in the documentation and the one most setup guides skip. It means your Bots are not sandboxed from each other. They are colleagues sharing one desk, one browser and one keychain.
The upside is real and it is the reason the design works: because the machine is shared, a Bot can hand a half-finished artifact to another Bot without re-uploading anything, and context survives the handoff. That is what makes a Chief of Staff pattern possible at all.
The consequence is equally real. A login you grant to one Bot is a login every Bot in your account can reach. You cannot give the Expense Manager access to your banking portal and rely on the Sales Outbound Bot being walled off from it. There is no per-Bot permission boundary to lean on.
So scope access at the account you sign in with, not at the Bot. If some system genuinely must stay separated, the boundary available to you is a different user, not a different Bot. The documentation does offer mitigations at the action layer rather than the identity layer: “Sensitive actions can go through Auto Review before they run,” data is encrypted in transit and at rest with training opt-out, and enterprise admins can set DLP, certificates, proxies and network controls at boot.
This maps directly onto the checkpoint question we covered in Human-in-the-Loop by Design. With a shared machine, your review gates are the main control you have, so place them on actions rather than trusting separation between agents.
Step 3: Learn the promotion ladder from Prompt to Routine
Most guides present Skills and Routines as two separate features to configure. The design documentation implies something cleaner. A Prompt “can be used once, saved as Skills, or triggered automatically as Routines.” One object, three stages.
Reading it that way turns setup into a progression rather than a configuration screen, and it tells you exactly when to promote something.
Table 2. The Prompt promotion ladder (CEOtudent editorial framework, built on SpaceXAI’s published object definitions)
| Stage | What it is | When to promote to this stage | What breaks if you promote too early |
|---|---|---|---|
| Prompt, used once | A one-off instruction in a chat | The task is genuinely new, or you are still discovering what good output looks like | Nothing. This is where exploration belongs |
| Skill, saved | The same instruction stored and reusable across any Bot with the right access | You have run it successfully more than once and the wording is stable | A half-correct Skill spreads to every Bot that calls it, so you multiply a mistake instead of a method |
| Routine, triggered | The workflow assigned to one Bot on a schedule or trigger | The output is reliable without correction, and the timing is genuinely predictable | The Bot runs a flawed workflow unattended, and you find out after several cycles rather than once |
The rule of thumb: prove it in a chat, save it as a Skill, schedule it as a Routine, in that order. Skipping straight to a Routine is how people end up with an agent confidently producing the wrong thing every Monday at 9am.
There is also a demonstration path. The product page describes it as: “Ask a Bot to follow along as you complete a workflow once. It saves it as a routine and runs it on its own next time.” Worth noting honestly: SpaceXAI’s own GTM guide describes the same recording feature as producing a skill rather than a routine, so the official wording is not consistent across pages. Whichever object it produces, the useful part is that you can teach by doing rather than by writing instructions.
If you are writing the instruction rather than recording it, the briefing structure in How to Delegate to an AI Agent transfers directly to a Skill, because a Skill is just a delegation brief you saved.
Step 4: Put things at the right level
The design documentation is explicit about what belongs where, and the reasoning is worth internalising because it predicts where your setup will get messy.
“Tools and Skills live at the account level because many Bots may need to browse the web, work with documents, or send email. Memory and Routines belong to the Bot because they reflect what that particular role knows and does over time.”
Table 3. Account level versus Bot level, and what it means in practice (CEOtudent analysis of the published architecture)
| Object | Lives at | Why | Practical consequence for your setup |
|---|---|---|---|
| Tools | Account | Many Bots need the same browsing, document and email access | Connect once, centrally. Adding a Bot does not mean reconnecting tools |
| Skills | Account | A method should be reusable by whichever Bot needs it | Write Skills generically. A Skill that hardcodes one Bot’s role will not travel |
| Memory | Bot | It reflects what that role has learned over time | Memory does not transfer. Splitting one busy Bot into two later means the new one starts naive |
| Routines | Bot | They are what that particular role does on a schedule | Routines are the thing you actually reassign when you reorganise your team |
The non-obvious implication is in the Memory row. Because Memory is per Bot and does not move, the cost of reorganising your Bots later is paid in lost context. If you expect a function to grow, it is cheaper to create the specialist Bot early and let it accumulate memory than to split an overloaded generalist in six months.
Step 5: Appoint a Chief of Staff, and know the ceiling
The Chief of Staff is a pattern, not a product feature. The design documentation records it as something users arrived at: “Some made a Chief of Staff Bot responsible for coordinating several specialists.” The GTM guide describes the same role concretely: a primary Bot that “owns meeting prep, inbox and post-call drafts, orchestrates the rest of the team,” and adds that “you can tell your chief of staff to spin up other agents, and organize your bots in sections.”
Bots coordinate in two documented ways. They can be placed together: “Put a few Bots in the same thread and they pass work between themselves.” And they can call each other: “your bots can call your other specialized agents.”
Two published numbers should shape how ambitious you get. The design documentation cites “practical limits of roughly 50 Bots per account and six per group chat.”
The group chat limit of six is the one that actually constrains design. It means a single coordinating thread holds a Chief of Staff plus roughly five specialists. Beyond that you are not running one team, you are running several that must hand off between threads. The account limit of about 50 is generous enough that most individuals will never reach it.
SpaceXAI’s own example Bots give a reasonable starting roster: Sales Outbound, Talent Scout, Paid Media, Expense Manager, Product Performance, Bug Reproduction, Account Health, and Chief of Staff.
A sensible first team, sized to the group chat limit:
- Chief of Staff. Owns your calendar, inbox triage and meeting prep. Routes anything a specialist owns.
- One revenue or output Bot for the work that generates your results.
- One admin Bot for expenses, scheduling and filing.
- One research Bot for background work that has no deadline.
- One domain specialist for whatever your job actually is.
Start with the Chief of Staff and one specialist. Add the rest when a real backlog justifies it, not in advance. The failure mode where people build an elaborate agent org chart before having work to give it is the one described in The Delegation Ceiling.
Step 6: A setup order that respects the architecture
Putting the four facts together gives a sequence:
- Choose the plan. Cursor Pro at $240 a year if you want agents only; SuperGrok at $360 if you also want the model and connectors.
- Decide the identity boundary first. Because isolation is per user, settle what this account may reach before connecting anything. Anything that must stay separate needs a separate user.
- Connect Tools once, at the account level. The GTM guide’s advice is to connect what you use daily rather than everything you can.
- Create the Chief of Staff and exactly one specialist. Two Bots is enough to learn the handoff.
- Work in chats for a week. Do not save anything yet. You are finding out which instructions are stable.
- Promote the stable ones to Skills. Write them so any Bot could run them.
- Promote only the reliable Skills to Routines. Put the review checkpoint on the action, not on the schedule.
- Add specialists when a backlog appears. Remember Memory does not transfer, so create a Bot early if you know the function will grow.
What this does not tell you
Two honest gaps. First, there is no published independent measurement of how well Grok Bot completes real work end to end; the product claims and example workflows here come from SpaceXAI’s own pages, which are marketing and documentation, not evaluation. Treat completion claims as vendor claims until third-party benchmarks exist.
Second, the pricing table is a subscription floor, not a total. Because usage beyond the weekly allocation is billed on token cost, the real annual number for a heavily used team of Bots is not knowable from the published prices, and we have not seen a public dataset of typical overage.
If you are deciding whether this belongs in your workflow at all rather than how to configure it, the prior question is which of your work is genuinely delegable, which we covered in Agent Literacy, and how your accountability changes once agents do the work, covered in Working With AI Agents Without Losing Accountability.
Frequently asked questions
Do I need SuperGrok to use Grok Bot?
No. Grok Bot is available to Cursor Pro, Pro+ and Ultra subscribers, to SuperGrok, SuperGrok Plus and Heavy subscribers, to Cursor Teams Standard and Premium, and to Enterprise. Cursor Pro at $20 a month is the cheapest listed plan that includes it, $120 a year less than SuperGrok.
Are my Bots isolated from each other?
No. Every Grok Bot shares one persistent cloud computer, including files, browser and logins, and the documentation states that isolation is per user rather than per Bot. Scope credentials accordingly.
What is the difference between a Skill and a Routine?
Both begin as a Prompt. A Skill is a Prompt you saved so it can be reused, and it lives at the account level so any Bot with the right access can run it. A Routine is a Prompt triggered automatically on a schedule, and it belongs to one specific Bot.
How many Bots can I have?
The design documentation cites practical limits of roughly 50 Bots per account and six per group chat. The six-per-thread limit is the one that shapes how you organise a coordinating team.
Is the Chief of Staff a built-in feature?
No, it is a usage pattern. It appears in SpaceXAI’s example Bot roster and its documentation describes users creating a Chief of Staff Bot to coordinate specialists, but you set it up yourself by creating a Bot and giving it the coordinating role.
Where does Grok Bot run?
On desktop, on macOS or Windows, and on iOS. Bots run on their own cloud computer, so they continue working when your own machine is closed.
Sources
SpaceXAI. Introducing Grok Bot, 11 August 2026.
SpaceXAI. Designing Grok Bot for a world of persistent agents, 3 September 2026.
SpaceXAI. Grok Bot product page, including plan listings and frequently asked questions, read 22 September 2026.
SpaceXAI. Pricing page, individual plans, read 22 September 2026.
SpaceXAI. Grok Bot for GTM, from the Grok Bot Guides series, read 22 September 2026.
Annual costs in Table 1 were calculated by us from the monthly prices published on the pages above. Table 2 is a CEOtudent editorial framework built on SpaceXAI’s published definitions of Prompts, Skills and Routines. Table 3 is our analysis of the account-level and Bot-level split described in the design documentation.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
