SpaceXAI launched Grok Bot on August 11, a set of AI teammates built to handle real work across applications, inboxes, and websites. Unlike a traditional chatbot that answers prompts, Grok Bot has its own computer in the cloud. It allows it to sign in into the tools users depend on and complete tasks from the beginning to the end. The bot only returns when an action calls for the user’s approval. Grok Bot can be messaged like a teammate from desktop or iOS, while conversations and work continue across devices.
It remembers the user’s pattern of working, learning from repetitive tasks without imparting similar instructions repeatedly. Multiple bots can work simultaneously and divide responsibilities across sales, marketing, operations, finance and engineering niches. These bots can work on one project while sharing context, with a group chat bringing several together and one coordinating the others. Bots can also pass assignments between themselves without the user managing each handoff. A user can demonstrate a task once and have the bot save it as a reusable routine.
What Grok Bot Costs and Who Can Access It
The product began as an internal prototype at SpaceXAI, where teams built bots for sales outbound, marketing campaigns, office operations, and bug fixes. SpaceXAI says a sales bot can research accounts, update CRM records, and create tailored follow-ups. Likewise, an operations bot can work on invoices, and an engineering bot can reproduce bugs and hand fixes to another bot. Grok Bot is in beta for SuperGrok Heavy, Cursor Ultra, and Cursor Teams Premium subscribers. It runs on macOS, Windows, Linux, and iOS, with Android listed as coming soon. Enterprise customers are directed to a waitlist for earlier access.
Introducing Grok Bot, now in early beta.
— Grok Bot (@bot) August 11, 2026
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
Musk’s newfound Bot access is currently bundled rather than sold separately. Cursor Ultra runs $200 per month for individuals, Cursor Teams Premium $120 per seat per month with centralised billing and SAML/OIDC sign-on, and SuperGrok Heavy $300 per month. With this, the Grok Bot is placed as an agent product inside subscriptions which customers already have paid for. The Cursor tiers appear because SpaceX acquired Anysphere, Cursor’s developer this year on June 16th for $60 billion in stock, which was also its largest startup purchase to date.
How Grok Bot Compares to Claude Cowork and Codex
With it’s launch Grok Bot has entered a highly competitive market where organizations move beyond conventional chatbots towards systems that can autonomously perform tasks. Anthropic’s Claude Cowork is similar in design around assigning longer tasks to an AI system. On the other hand, OpenAI has placed Codex around completing work and software-centric tasks rather than generating plain responses. Microsoft is developing Scout as an always-on personal agent capable of working across Microsoft 365 and other ecosystems.
Google’s Gemini Spark will also move towards AI that can autonomously operate on a user’s behalf. Differentiation between these products comes down to how much autonomy they get, what applications they can access, and how much oversight is necessary. Grok Bot’s approach is focused on giving each bot its own computer and allowing it to work across websites and applications, even when the services lack a dedicated API.
That makes it well-suited for workflows that conventional AI agents struggle to acquire. The direction is consistent across vendors, moving from following instructions to executing autonomously. Grok Bot is AI’s latest attempt to compete in that sphere.
How Safe Is Grok Bot’s Memory and Data Handling?
Grok Bot’s capabilities come hand-in-hand with certain questions that the launch does not answer. SpaceXAI says bots retain information about a user’s working style, but the announcement does not say where that context is being stored, or how long it remains on-device, or whether users can delete or audit that context. Data acquisition is yet another concern. Bots can log into Gmail, CRM systems and internal operational mechanisms, which means handling confidential information and customer information at the of completing assigned tasks. And this concern is not hypothetical at all. Earlier this year, users reported agentic models deleting files and databases without authorisation, and OpenAI disclosed that two of its models escaped a test environment and reached a third party’s production systems.
The announcement explains what bots can access and do, but does not explain what information SpaceXAI itself can see during the processes, how the data is preserved, or what controls businesses have over that access. The safety model also remains an important yet unanswered question. SpaceXAI shares that the bot returns to the users when something requires a mandate or an approval. But the announcement does not specify which actions automatically require human confirmation and how permissions are managed for sensitive operations.
There is also the question of what happens when a bot gets something wrong. A mistake in a draft can be corrected, but an incorrect email sent to a customer, a wrongly modified CRM record, or an irreversible action could have unforeseen consequences before any human oversight. The announcement does not mention details on rollback mechanisms, recovery procedures, or transparency when a bot makes an irreversible mistake. That matters because an AI teammate depends not only on how much work it can complete, but also whether users can safely authorize that work.
The more permissions a bot receives, the more crucial its permission controls, audit trails, and human oversight mechanisms become. Grok Bot therefore arrives with an intent to instill confidence in users to hand over their work and let an AI teammate finish it. But turning that promise into a reliable workplace tool will depend on whether SpaceXAI can provide visibility and control around the bot’s memory, data access, and autonomy.
You may also like to read – Grok 1.0 to 4.3: What Improved and What Still Lags
