Grok Bot Arrives: An Always-On AI Teammate With Its Own Cloud Computer

Introduction

Grok Bot has moved into beta, giving SpaceXAI a new kind of AI product: not just a chatbot, but a persistent AI teammate designed to carry out real work inside the tools people already use.

The original report described the launch as a major step toward agents that can behave more like remote coworkers. That framing is broadly consistent with SpaceXAI's official product description. Grok Bot can work from a persistent cloud computer, sign in to apps and websites, use browser-based tools, keep working while the user is away, and return when a task is complete or needs approval.

At the time of the original article, Grok 4.6 was still being discussed as an imminent release. SpaceXAI subsequently launched Grok 4.6 on August 13, 2026, with availability in Cursor, Grok Build, the SpaceXAI API, and selected partner platforms.

A new model and a new agent product arriving within days of each other makes the direction clear: SpaceXAI is pushing beyond conversational AI toward agents that can stay active, use software directly, and complete multi-step workflows.

AI Teammate on the Job, Even When Your Laptop Is Closed

Grok Bot is positioned less like a conventional assistant and more like a teammate you can hand work to.

According to SpaceXAI's documentation, Grok Bot uses a persistent cloud computer with a browser, filesystem, and terminal. Bots can work with connectors and MCP integrations where available, while also using the browser and computer interface for websites that do not expose a clean API.

That changes the workflow in an important way. Instead of generating instructions for you to carry out manually, the Bot can work inside the actual application.

For example, you can sign in to a service such as Gmail or a business application on the shared cloud computer, then let the Bot continue from that authenticated session. When a password, two-factor code, CAPTCHA, or another

A human-only step is required, and the computer can be handed back to you so you can complete the sensitive step before returning control.

This cloud-based design also means a task does not have to stop just because you close your laptop. Background turns and scheduled routines can continue on the cloud computer.

That is one of the biggest differences from local-only agent workflows, which depend on the user's machine remaining online.

Multiple Bots in Parallel: One Person, a Small AI Team

A single Bot is useful, but Grok Bot is designed around the idea that several specialized Bots can work together.

You might create one Bot for inbox triage, another for sales research, another for competitive monitoring, and another for drafting content. SpaceXAI says Bots can run in parallel, message each other, share context through conversations and files, and hand off work.

There is an important implementation detail: Bots on the same account share one persistent cloud computer. That means they can share files, browser sessions, and logins, while each Bot gets its own screen for computer-use tasks. Separate Bots should therefore be treated as different roles, not as separate security boundaries.

A community post highlighted in the original article collected a large number of possible Grok Bot workflows, ranging from outbound sales to reporting and inbox automation.

Early user Lenny Rachitsky also shared several practical examples, including matching job seekers with hiring companies, helping with support email, reviewing card statements for recurring subscriptions, and preparing research briefs for podcast guests.

The same idea extends to more everyday workflows. One example shown in the source article used a Bot to compare grocery options across online stores before preparing a cart.

Another example showed the Bot helping compare food options and navigate an ordering workflow.

![Image showing a conversation interface where the user asks whether there are high-protein options, preferring simple starches and low fat. The Bot replies that it prefers simple items, not loaded bowls, and opens the Sweetgreen cart. It then asks the user which one they should choose, and the Bot lists simple options with real prices in Fredrika's Sweetgreen cart, such as Kale Caesar, Classic Chicken Caesar wrap, and others. The user says they will skip Harvest and Pesto Farm, and the Bot suggests the Gai Chicken Rice, whose auto-order time is 5:15. The Bot also introduces the store's simple options, such as The Original Gai, and pulls up prices.](https://we0-cms.oss-cn-beijing.aliyuncs.

The broader point is not that every purchasing or external action should be automated without supervision. SpaceXAI's own documentation recommends keeping sensitive or consequential actions—such as sending, purchasing, deleting, publishing, or changing production systems—behind explicit approval boundaries.

Teach the Bot by Showing It Once

One of Grok Bot's more interesting features is the ability to turn a demonstrated workflow into a reusable process.

Instead of describing every browser action in a long prompt, users can use the "Teach a task" flow when it is available. The basic process is straightforward:

1. Open a one-to-one conversation with a Bot and open its computer view.
2. Choose Teach a task.
3. Describe the result you are about to demonstrate.
4. Perform the browser workflow once.
5. Stop the recording and review the draft skill the Bot creates.
6. Test the skill on a safe example before turning it into a routine.

The learned workflow is not meant to be treated as perfect after a single demonstration. SpaceXAI recommends reviewing the resulting skill, adding decision rules and failure handling, and defining which actions require human approval.

Once the workflow is reliable, it can be turned into a routine. A routine can run on a schedule and, where supported, can also be triggered by an event.

That removes a lot of the setup normally associated with personal automation. For many routine workflows, users may no longer need to manually build a separate cron job, custom browser automation stack, Docker environment, or MCP service just to keep a recurring task running.

The laptop can be closed while a background routine continues on the cloud computer.

Cursor Integration Gives Grok Bot a Strong Starting Point

The original article framed Grok Bot as a major product integration between SpaceXAI and Cursor. That part is clearly visible in the official product experience.

Grok Bot uses Cursor authentication, and eligible access currently includes SuperGrok Heavy, Cursor Ultra, and Cursor Teams Premium. Cursor team settings can also affect Grok Bot, including privacy settings, MCP configuration, team rules, and Cloud Agent access.

For team environments, administrators can control whether Bots are allowed to launch Cursor Cloud Agents. SpaceXAI's documentation says that setting is enabled by default for supported teams.

The pricing shown on the official Grok Bot page currently lists:

Plan	Price	Grok Bot Access
Cursor Ultra	$200/month	Included
Cursor Premium Teams	$120/seat/month	Included for eligible premium seats
SuperGrok Heavy	Check current Grok pricing	Included for eligible subscribers

Cursor Premium Teams also adds centralized team billing and settings, a

team marketplace for skills and plugins, shared usage analytics, and SAML/OIDC SSO.

Pricing and usage limits can change, so the official Grok Bot access page and Cursor pricing documentation should be treated as the current source of truth.

What About the Reported Cursor Acquisition?

The original Chinese article states that Elon Musk had acquired Cursor for $60 billion. I could not find an official SpaceXAI or Cursor announcement confirming that claim.

Official Cursor materials instead describe a partnership with SpaceX to use SpaceXAI's Colossus infrastructure for model training, while SpaceXAI and Cursor now share deep product integrations around Grok models and Grok Bot.

For that reason, this publishable version does not present the reported $60 billion acquisition as a confirmed fact.

Grok Bot vs. Claude Cowork

The original article also compared Grok Bot with Anthropic's Claude Cowork, and the comparison is understandable.

Both products are moving toward the same broad destination: AI systems that can take on multi-step work, interact with tools, and continue working remotely rather than remaining tied to a single chat window.

Anthropic says Claude Cowork can run on web, desktop, and mobile, and its remote sessions can continue in the cloud after the user closes a laptop. Grok Bot similarly keeps work running on its cloud computer and emphasizes persistent named Bots, parallel coordination, shared files and sessions, and repeatable routines.

The distinction is less about whether one product is an "agent" and the other is not. The more useful question is how each product handles persistent context, computer use, integrations, approvals, team administration, and recurring work.

Grok Bot's integration with Cursor is particularly relevant for users already working in the Cursor ecosystem, while Claude Cowork is positioned as a broader knowledge-work environment across files, browser tasks, and other productivity workflows.

Why Grok Bot Matters

The significance of Grok Bot is not simply that another AI agent has launched.

It represents a shift from an assistant that waits for prompts toward a persistent worker that can remain active, use real applications, maintain context, and return with completed work. The interaction model also becomes closer to delegating a job than writing a perfect prompt.

That does not remove the need for human oversight. In fact, the more capable an agent becomes at taking actions inside real systems, the more important approval rules, account permissions, privacy settings, and auditability become.

But the direction is becoming clear: AI products are moving from "answer this question" toward "take ownership of this workflow and come back when it is done."

FAQ

What is

Grok Bot?

Grok Bot is a cloud-based AI teammate from SpaceXAI. It can use a persistent cloud computer, work inside apps and websites, manage files, run terminal tasks, and coordinate with other Bots instead of only answering in chat.

Does Grok Bot keep working when my laptop is closed?

Yes. SpaceXAI states that Bot work runs on the cloud computer, so closing the desktop app, laptop, or iPhone does not stop a background task or routine that is already running.

Can several Grok Bots work at the same time?

Yes. Multiple Bots can work in parallel and coordinate with one another. Bots on the same account share the same persistent cloud computer, including its files, browser sessions, and logins, so they should not be treated as isolated security environments.

Can Grok Bot use websites without an API or MCP integration?

In many cases, yes. Grok Bot can use browser-based computer interaction for services that do not offer a clean API or MCP connector, although individual websites may block automation, require a login, show a CAPTCHA, or request human confirmation.

Can I teach Grok Bot a workflow by demonstration?

Yes, when the "Teach a task" feature is available for your account. You can demonstrate a browser workflow once, review the generated draft skill, test it, and then turn it into a repeatable routine if it behaves reliably.

What actions should still require human approval?

Sensitive actions such as sending external messages, making purchases, deleting data, publishing content, changing production systems, entering passwords, or completing two-factor authentication should remain behind explicit approval or takeover steps. SpaceXAI's documentation recommends defining these boundaries before automating recurring work.

How much does Grok Bot cost?

Grok Bot is available through eligible plans including Cursor Ultra, Cursor Teams Premium, and SuperGrok Heavy. Cursor Ultra is listed at $200 per month and Cursor Premium Teams at $120 per seat per month on the official Grok Bot page, but pricing and usage allowances can change.

Is Grok 4.6 available in Cursor?

Yes. SpaceXAI released Grok 4.6 on August 13, 2026, and lists Cursor as one of its launch environments. Cursor's model documentation also lists Grok 4.6 and Grok 4.6 Fast.

Related Tools

• Grok Bot: SpaceXAI's persistent cloud AI teammate for multi-step work across apps, websites, files, and routines.
• Cursor: An AI coding environment with agent workflows, Cloud Agents, plugins, MCP support, and Grok model integration.
• Grok 4.6: SpaceXAI's current flagship model for coding, knowledge work, and agentic tool use.
• Claude Cowork: Anthropic's agentic workspace for multi-step knowledge work across files, tools, browsers, and cloud sessions.

Related Links

• Introducing Grok Bot: SpaceXAI's official launch announcement and product overview.
• Grok Bot Documentation: Official explanation of persistent cloud computers, multi-Bot coordination, memory, and

workflow behavior.

• Grok Bot Get Started Guide: Installation, authentication, supported platforms, and first-run setup.
• Skills and Routines: Official guide to reusable skills, teach-by-demonstration, schedules, and event-triggered routines.
• Grok Bot FAQ: Official answers about cloud execution, approvals, privacy, pricing, supported platforms, and shared computers.
• Cursor Models and Pricing: Current Cursor model availability, Grok 4.6 pricing, and plan details.
• Cursor and SpaceX Model Training Partnership: Cursor's official announcement about using SpaceXAI's Colossus infrastructure for model training.
• Claude Cowork: Anthropic's official Cowork product page for comparison with cloud-based agent workflows.

Summary

Grok Bot brings the AI-agent idea closer to a persistent digital teammate. It can work from a cloud computer, use real apps and websites, continue after your laptop is closed, coordinate with other Bots, and turn successful workflows into reusable skills and routines.

Its close integration with Cursor gives it a practical starting point for developers and teams already using Cursor's tools, while its browser-based computer use opens the door to workflows that do not have clean APIs.

The most important limitation is also the most important design principle: giving an AI agent real access to accounts and applications makes approval rules, privacy settings, and security boundaries essential.

The real shift is from asking AI for an answer to handing AI a workflow and expecting finished work back.