We need bots, but do they need us? Not a whole lot, if Grok Bot can do everything it purportedly can. SpaceXAI introduced the new AI agent platform this week and says it can do it all — message each other, do group chats, share work and even answer to an AI boss.
Grok Bot was jointly developed by SpaceXAI and Cursor, an AI development company that SpaceX is acquiring for $60 billion. Grok Bot is not to be confused with Grok, which is an AI chatbot that can answer questions and generate images, videos and code. Grok Bot is designed to be a team of AI agents that can help with work tasks and projects.
AI agents differ from AI assistants in that they can work autonomously and proactively to execute projects. Assistants can make dinner reservations for you; agents can help run a restaurant. Professional services firm Deloitte estimates that 75% of companies plan to incorporate AI agent systems into their operations by 2028.
Grok Bot is only the latest AI agent to come along. Predecessors include OpenClaw, Sai by Simular, Claude Computer Use, Manus, OpenAI Operator, NanoClaw and Hermes Agent. These options vary by what they can do, how secure they are and how complicated they are to use.
How does Grok Bot work?
Customers can create one or more AI agents that use their own cloud-based computer — not the customer’s — and can access the apps, websites and email inboxes the customer so designates. The agents can execute jobs from beginning to end and will contact the customer only if an approval is needed, SpaceXAI said.
Customers can message the AI agents by phone or computer to hand off a project or discuss an ongoing one. The company said Grok Bot can “remember conversations, learn how you like things done and get sharper the more you work together.”
SpaceXAI said Grok Bot doesn’t require the customer to set up workflows or routines — “simply message a Bot to take on a task, and it gets it done.”
What can Grok Bot do?
In its announcement, SpaceXAI said employees have been using Grok Bot in-house. The company said that bots have helped with sales, recruiting, processing invoices and implementing bug fixes.
SpaceXAI said Grok Bot agents can work in concert, with each handling a separate operation — email, expenses, recruiting, bug fixes and operations. The agents can send messages to each other, pass work off if need be and assign ownership of tasks.
Customers can also create an agent to act as “chief of staff” to manage the other agents.
Learning from the customer
SpaceXAI said customers can train Grok Bot agents by having them follow along and watch their workflow to learn the steps and processes the customer uses. After learning the routines, the agents can fly solo the next time “without re-explaining it again,” the company said.
Grok Bot becomes more proactive and self-directed over time, completing necessary work before the customer even requests it, SpaceXAI says.
SpaceXAI said Grok Bot — currently in beta — is now available to subscribers of SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium on desktop and iOS. Enterprise users can join a waitlist for future access. After downloading the app, customers can create a Grok Bot teammate and then assign tasks and projects.
Trust is paramount
Grok Bot, like other AI agents, must be reliable from start to finish, or else trust in these “co-workers” will disappear, says Uttara M. Ananthakrishnan, an assistant professor in the Foster School of Business at the University of Washington.
“The biggest issue pops up with multi-step jobs,” Ananthakrishnan told CNET. “If a bot gets through the first two steps of a process and then flakes out, the person using it has to start all over. If a task needs approvals from five different people in different departments, a bot is going to be terrible at it.”
Ananthakrishnan said it’s a “game changer” if companies can seamlessly integrate AI agents into their operations to do repetitive tasks, save time and increase efficiency. But if the AI agents are clunky, workers will revert to their own ways of doing things.
“People will always have their own optimized workflows and won’t tolerate an agent making minor errors. If they try two times and if it doesn’t work, they will just roll their eyes and think about it as an AI for AI’s sake product,” Ananthakrishnan told CNET.
Going rogue
In theory, AI agents like Grok Bot, OpenClaw and others can be very useful for companies. But sometimes they can be too self-directive. Their abilities can sometimes go off the rails.
In a recent instance in Australia, a man wanted OpenClaw to book him a spot in an exercise class. There was a waiting list, but the AI agent managed to move him up the list — by deleting someone else who was ahead of him. OpenAI and Anthropic also disclosed last month that their AI models hacked other companies’ servers during internal cybersecurity tests.
And in another incident, Anthropic’s top AI model used fake identities to deceive humans and insert malicious code during testing by Britain’s AI Security Institute.