Always-on AI agents: Grok Bot, ChatGPT dots and Claude Cowork

What the new agents that keep working for you actually are, who can get them, and how to use one without losing control.

A new kind of AI tool arrived this autumn. Instead of answering one question and stopping, these agents take a goal and keep working on it. Some keep going around the clock. That can save you real time, and it also means the rule the AI drafts, you send matters more than ever.

What they are, in plain words

Think of an assistant with their own laptop. You give them a goal, they sign into the apps you allow, and they work through the task without you watching. When they're done, they bring you the results to review.

• Grok Bot from SpaceXAI is described as a team of always-on agents. Each one has its own computer, works inside tools and apps like you do, and keeps working 24/7. It's included with SuperGrok, Cursor Pro and Cursor Teams plans, and there's a version for Enterprise. Team Bots are shared teammates for a whole team.
• Dots in ChatGPT are always-on agents with their own cloud computer. You give a goal, connect apps and decide what it may do on its own. Results come back for your review. They're rolling out to eligible Pro and Business Premium users aged 18 and over, and at launch not to Pro users in the EEA, the UK or Switzerland.
• Claude Cowork works a little differently. Anthropic has made Cowork available from any conversation, without picking a separate mode. It's rolling out to Pro and Max plans, and your Projects carry over.

Why the rule matters more now

When you chat with an AI, you see every answer as it appears. An agent that works while you sleep can take dozens of steps you never see. One wrong email, payment or post can happen at 3am. So keep the last step for yourself. The agent prepares, you approve.

How to use one safely

1. Start with read-only tasks. "Read my inbox and list the customers waiting on a reply." "Check my unpaid invoices and summarize them." Nothing is changed, so a mistake costs you nothing.
2. Keep sending, paying and posting behind your approval. Set the agent to ask before any of these, every time. Draft emails are fine. Sent emails are yours to send.
3. Connect only the apps the task needs. An agent for invoices doesn't need your social media accounts.
4. Review connected-app permissions monthly. Disconnect anything you no longer use, in the agent and in the apps themselves.
5. Read the results before acting on them. Treat the agent's report like a new assistant's notes: useful, and worth checking.

One more thing worth knowing. SpaceXAI's own Grok Bot docs say that all of your Bots share one cloud computer, and that you should not use separate Bots as a security boundary. In plain words: splitting work across several Bots doesn't keep it apart. Keep sensitive data out of any agent that doesn't need it.

A prompt to start with

Copy this as your first task for any always-on agent:

Your job this week is read-only. Read my inbox and my unpaid invoices list. Do not send, reply, pay, post, delete or change anything. Each morning, give me a short list: customers waiting on a reply, invoices overdue, and anything that looks urgent. If a task would need you to send or change something, stop and ask me first.

Once the reports are reliably right, let it save drafts too. The sending stays with you, as in let AI draft every reply, never send one.

• More guides
• Let AI draft every reply, never send one →
• Write rules once so Claude remembers your voice →
• Why every AI draft needs a second agent checking it →