SpaceXAI (formerly xAI) launched the early beta of Grok Bot today, shifting its product positioning from chatbot to “an AI teammate that can take over work.” Each Bot comes with its own dedicated cloud computer, allowing it to log into users’ existing tools and websites like a real person, executing tasks around the clock and only checking back with humans when approval is needed. This is SpaceXAI’s latest move in catching up with Anthropic and OpenAI on the AI Agent track. SpaceXAI has been busy recently—just last month in June, it signed a merger agreement with AI coding tool Cursor, with an implied equity value of $60 billion (approximately NT$1.93 trillion), expected to close in Q3 (possibly within the next few days).
Core feature: have your own cloud computer.
The biggest difference between Grok Bot and typical chatbots is that it “has its own computer.” Each user gets a continuously running cloud computer, which the Bot can use to access files, browsers, websites, and applications. Even when the user shuts down their laptop, the cloud work continues.
Introducing Grok Bot, now in early beta.
Bots are AI teammates that do real work for you. They sign in to your tools, use them just like you do, and come back with finished work. pic.twitter.com/uyfA97yo98
— Grok Bot (@bot) August 11, 2026
The company emphasized that the Bot operates exactly like a real person, logging directly into accounts and taking actions on-screen. This allows it to handle platforms that have no API and are not connected to MCP (a universal interface standard for AI to connect external tools). Roman of the product team said in the announcement that most AIs can only get to about “90% done”; the difference with Grok Bot is that the work ends up where a human would naturally put it—in the actual tools.
The internal cases cited in the announcement include: a sales Bot that writes call records into CRM and drafts follow-up emails, an operations Bot that assigns seats to new employees and processes invoices received via Gmail, and an engineering Bot that reproduces product interface errors, creates tickets, and then hands the fix to another Bot dedicated to debugging.
Multi-Bot Collaboration: Chief of Staff Plus Dedicated Bots
Grok Bot’s scheduling uses a “multiple bots in parallel” architecture. Users can create a “chief of staff” bot to coordinate tasks, with dedicated bots underneath for inbox, expenses, recruiting, or debugging. The bots can message each other directly, share context from the same project, and can also be placed in a group chat to delegate work among themselves.
The way to teach a Bot to do something is to ask it to “watch along the next time you do the same work,” and the Bot saves the steps as a repeatable process (officially called a Routine) that it automatically runs the next time. This is much easier to get started with than having users draw a flowchart first and configure every automation node. Emma from the operations team recalls in the official announcement that at first she checked the Bot’s progress every 15 minutes, watching it so closely that the Bot asked her why she kept asking, before she gradually learned to let go.
Subscription Plans and Pricing
As of August 12, the Grok Bot beta is available to users of the following subscription plans:
- SuperGrok Heavy: Under the existing subscription plan, Grok Bot is already included.
- Cursor Ultra$200 per month (approximately NT$6,500), including cloud computer, login tools, scheduled routines, desktop and mobile device access, and a higher AI token allowance.
- Cursor Premium Teams$120 per seat per month (approximately NT$3,900), with Ultra features plus team billing, Skills/Plugins marketplace, and SAML/OIDC single sign-on.
Even Grok free users can choose the Free trial plan for a free trial after installing the desktop version (credit card information required), though the token allowance is quite limited. Subscription plans include weekly usage, and any overage is charged separately based on token cost.
Support platform
Grok Bot currently supports the following platforms:
- macOSApple Silicon and Intel
- Windows：Windows 10/11
- iOS：iPhone
- Linux：Debian/Ubuntu
- AndroidComing soon
Enterprise customers, however, need to join a waiting list, and it has not been fully opened yet.
Security Risk: Shared Cloud Computers
There is a noteworthy security concern in Grok Bot’s design. The official FAQ clearly states that all bots created by the same user share one persistent cloud computer, including files, browser, and login state. The isolation mechanism operates between different users, not between different bots.
This design lets the sales bot hand off data to the follow-up bot, and also removes the hassle of each bot logging in repeatedly. However, if one bot gains inappropriate file or account permissions, the impact can spread to the other bots. For enterprises that require strict permission segregation, it cannot be assumed that “one bot per role” means the data is already isolated.
xAI said that data in transit and at rest on the cloud computer is encrypted, users can choose not to have their data used for training, and sensitive actions can also be checked by the Auto Review mechanism first. Enterprise plans can also configure data loss prevention, certificates, and network controls. However, Grok Bot is still in beta, and the official team has not publicly disclosed error rates, Auto Review’s interception effectiveness, or third-party security assessments.
Conclusion
Grok Bot’s product direction is clear: it doesn’t just generate answers—it goes into the actual websites and applications to get the work done. With its proprietary cloud computer, Routine workflow learning, multi-bot collaboration, and continuous memory, it’s closer to a digital colleague than a typical chatbot.
However, once AI can log into accounts and modify data, evaluation criteria can no longer be based solely on response quality. Until independent reliability data is available in the beta phase and the permission boundaries of shared cloud machines are clarified, Grok Bot recommends starting with low-risk, auditable, and recoverable processes.
Source: KOCPC Chinese