AI Pulse · Under the Radar

Grok Bot went viral this week because it gives each AI agent its own persistent cloud computer. The idea is simple. Instead of asking an assistant a one-off question, you can assign work to an agent that has a workspace, tools and some continuity between tasks.

Within days, builders released useAgent, an AGPL open-source clone with a similar model. It creates isolated Linux sandboxes for individual agents and connects them to tools many teams already use, including Slack, GitHub, Gmail and Linear. It also includes versioned skills and human approval steps, so a person can check an action before the agent takes it. Rather than charging through a separate metered product, useAgent is designed to run on a company’s existing Claude Code or Codex subscription. The project has reached 269 GitHub stars, reportedly growing at 92 per day.

For an owner, the important point isn’t which project wins. It’s how quickly a popular AI product idea can now become available in open source. That can reduce dependence on one vendor and give technical teams more control over where agent work runs and what systems it can access. It also raises the bar for governance. Giving an agent access to email, code repositories and task systems needs clear permissions, approval rules and an audit trail.

This is the kind of thing we build into an AI command centre, where agents can be useful without becoming another unmanaged system with access to your business.