From a Restricted Beta to More Plans in Ten Days

xAI announced on August 21, 2026 the expansion of Grok Bot beyond its initial beta, launched just ten days earlier on August 11, according to the company's official announcement confirmed by 9to5Mac and VentureBeat. The agent, previously limited to a small group of subscribers, became available on the SuperGrok Plus, SuperGrok Heavy, Cursor Pro+, Cursor Ultra and Cursor Teams plans.

Its Own Cloud Computer for Every Agent

According to xAI's official announcement page, each Grok Bot runs on its own cloud computer, equipped with a browser, filesystem and terminal. The agent can sign into the tools the user already relies on, using that person's own credentials, and work through multi-step tasks autonomously while the user is away.

Surfaces Only When It Needs Approval

According to VentureBeat, which describes Grok Bot as a persistent digital coworker, the agent only draws the user's attention when a specific decision requires human approval, functioning somewhat like an employee who checks in only when they hit a real obstacle while carrying out a task.

Pricing by Plan: From $200 to $120 per Seat

According to Blockchain.News and 9to5Mac, the Cursor Ultra plan, at $200 a month, includes Grok Bot's own computer, sign-ins to the user's tools, scheduled routines, and desktop and mobile access with expanded token limits. Cursor Teams Premium, at $120 per seat per month, adds centralized billing, a team agent marketplace, shared usage analytics, and single sign-on via SAML or OIDC.

Why It Matters for Brazilian Agencies and SMBs

Expanding Grok Bot to more accessible plans, including SuperGrok Plus and Cursor Pro+, puts AI agents capable of autonomously operating tools within reach of smaller teams, not just large corporate accounts. For Brazilian agencies managing multiple client tools, such as ad dashboards, spreadsheets and service systems, the pitch of an agent that works alone through multi-step tasks and only interrupts the team when necessary could cut time spent on repetitive work, as long as the team clearly defines which actions require human approval before granting broad access to clients' own credentials.