xAI, the Elon Musk owned AI company now folded into SpaceX, introduced a new AI agent platform called Grok Bot this week, positioning it as a team of always on digital coworkers rather than a chatbot.
Grok Bot was built jointly with Cursor, the AI coding company SpaceX is acquiring for $60 billion, according to CNET. The company said Grok Bot is separate from Grok, the chatbot that answers questions and generates images and video. Grok Bot instead runs autonomous agents meant to execute entire projects rather than single tasks.
Each agent operates on its own cloud based computer, not the customer’s device, and can access the apps, websites and email inboxes a customer designates. SpaceXAI said the agents work through jobs from start to finish and only contact the customer when they need approval to proceed.
Customers can message an agent by phone or computer to hand off work or check on progress. The company said agents remember past conversations and get better at matching a customer’s preferences the more they are used, and that setting one up does not require building workflows first.
Multiple agents can work together, according to the announcement, passing tasks to each other and dividing work such as email, expenses, recruiting and bug fixes. Customers can also create one agent to act as a chief of staff overseeing the others.
Grok Bot is currently in beta, available to subscribers of SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium on desktop and iOS, with enterprise access limited to a waitlist for now.
Uttara M. Ananthakrishnan, an assistant professor at the University of Washington’s Foster School of Business, told CNET that trust will decide whether tools like Grok Bot stick. She said multi step jobs are the biggest risk, since a bot that fails partway through a process forces the user to start over, and tasks needing sign off from several people in different departments remain a weak spot for current agents.
Grok Bot joins a growing field of AI agents that includes OpenClaw, Manus, OpenAI’s Operator and Anthropic’s Claude Computer Use, each with different strengths and different track records. Some of those agents have run into trouble acting too independently, including one case in Australia where an agent moved a user up an exercise class waitlist by deleting another person who was ahead of them.
