Grok Bot: AI coworkers that work in your apps around the clock

On August 11, 2026, xAI launched the beta version of Grok Bot. It is a platform of so-called agents, meaning AI programs that carry out tasks themselves rather than just answering questions. Each agent gets a name and works as a kind of digital coworker.

Users message the bots the way they would message a colleague. According to xAI, each bot has its own computer in the cloud, can work around the clock and handles multistep tasks from start to finish.

Logs in like a human

The bots log in to the user's tools and operate them the way a person would. They work across apps, websites and files. According to xAI, they only come back with the work when human judgment is required.

xAI also says a bot can learn a workflow by watching it once and then repeat it. Multiple bots can work at the same time and talk to each other. These are xAI's own claims and have not been independently verified.

As examples of use, xAI mentions overnight research, drafts of emails and LinkedIn messages, cleanup of the CRM sales system, onboarding of new employees, invoice processing, filing bug reports, marketing campaigns and coordination of office work.

Pricing and availability

Grok Bot has its own usage quota, separate from existing Grok and Cursor subscriptions. Delegated work therefore does not eat into regular subscription credits.

According to VentureBeat, access costs $200 per month for individuals (Cursor Ultra) and $120 per seat per month for teams (Cursor Premium Teams). It is also included in SuperGrok Heavy at $300 per month. xAI itself mentions other subscription tiers, however, so it is a good idea to check what currently applies.

The beta is available for macOS, Windows, Linux and iOS, with Android coming soon. Companies can join a waitlist.

Caveats

VentureBeat points out that agents operating systems in daily use, such as CRM records and support queues, can cause operational problems if they fail. According to the article, success depends on permissions, predictable execution, escalation to humans and accurate memory.

SpaceXAI has not published benchmarks, meaning standardized performance tests. Model selection happens automatically and cannot be controlled by the user, which one tester has criticized.

Early users have been positive, though. Lenny Rachitsky called it the most exciting AI product in a while. Matt Shumer highlighted the product's orchestration but said the automatic model router was not good during his testing.