이미지: X — 뉴스 앰프
Summary
- xAI has released a new agent called Grok Bot in early beta for Grok Heavy, Cursor Ultra, and Cursor Team Premium users
- Users build and run their own bots, which operate their own dedicated computers to carry out real tasks
- It was rolled out simultaneously in the coding tool Cursor, hinting at a possible link between xAI and Cursor
- 신제품
- Grok Bot 에이전트
- 공개 시점
- 2026년 8월 11일, 초기 베타
- 대상 이용자
- Grok Heavy, Cursor Ultra, Cursor Team Premium
- 작동 방식
- 봇 전용 클라우드 컴퓨터 24시간 구동, 루틴 저장·봇 간 협업
- 지원 플랫폼
- macOS·Windows 10/11·iOS·Linux(Debian·Ubuntu), Android 예정
- 공식 활용 예시
- 영업·채용·광고 운영·경비·제품 분석·버그 재현·계정 점검·비서실장 8종
What was released
xAI has released a new agent called "Grok Bot" in early beta for Grok Heavy, Cursor Ultra, and Cursor Team Premium users. The news first spread via a news amplification account on August 11, 2026, and xAI simultaneously published product details on its official page. Users can build and run their own custom bots, and rather than simply spitting out answers in a chat window, the bots operate a dedicated computer to carry out real tasks. The same feature was rolled out at the same time to Cursor's paid tiers (Ultra, Team Premium), a coding tool.
Specs confirmed on the official page
Each bot runs on its own dedicated computer in the cloud. Even if a user closes their laptop, the bot keeps working around the clock, and multiple bots can be run in parallel to handle tasks simultaneously. Bots can log into various services on the user's behalf to operate tools, and once a task is demonstrated once, the process can be saved as a "routine" for repeated execution. Bots can also collaborate by handing off tasks to one another — for example, assigning research to a research bot, then passing the results to a communication bot.
| Item | Details |
|---|---|
| Operation | Dedicated cloud computer per bot · runs 24/7 · multiple bots in parallel |
| Supported platforms | macOS (Apple Silicon, Intel), Windows 10/11, iOS, Linux (Debian, Ubuntu), Android coming soon |
| Repetitive tasks | Save a demonstrated task as a "routine" for re-execution |
| Collaboration/memory | Task handoff between bots, retains conversational context and preferences |
| Usage requirement | Included with SuperGrok Heavy or Cursor Ultra subscriptions |
The official page lists eight example use cases: sales outreach, talent sourcing, paid ad management, expense management, product performance analysis, bug reproduction, account health checks, and a "Chief of Staff" role overseeing schedules and tasks. Notably, more of the examples relate to general office and operations work than to coding.
What this means
By 2026, the term "AI agent" has become the standard label for the next stage beyond chatbots. Rather than just answering questions, it refers to software that directly operates a computer screen — opening files, clicking through websites, running code — to carry out real tasks on a user's behalf. Google's "Gemini Spark," which launched in Korea on July 30, is a prominent example. Spark is a personal agent that runs 24/7 on Google Cloud, moving between Gmail and documents and continuing to work even when a laptop is closed. Microsoft also published a guide on its official blog on August 10 explaining how to build agents using only natural-language descriptions, without writing a single line of code.
What stands out about Grok Bot is that xAI has embedded this trend not only in its own Grok chatbot ecosystem but simultaneously in Cursor, a developer-focused editor. Cursor is an AI coding assistant widely used by developers to write and edit code. Attaching xAI's agent to it suggests the scenario could expand to bots directly executing coding tasks themselves.
So what changes
Until now, developers using Cursor to write code had to keep feeding prompts manually. If Grok Bot is genuinely available at this tier, it points toward usage where users set up a bot once and hand off repetitive tasks — running tests, organizing files, monitoring for certain conditions — to it. As the eight examples on the official page show, the intended scope extends beyond development to general business functions like sales, recruiting, and expense processing. The chatbot competition has shifted from "smarter answers" to "software that works on your behalf," and the model of agents each having their own dedicated computer working continuously in the cloud is quickly becoming reality. Still, since this is an early beta, real user feedback and stability testing have yet to accumulate.
