A Chinese startup built AI employees that worked so hard the humans had to hide from them in a private Slack channel.

The Summary

  • Xiankun Wu, CEO of Kuse, deployed OpenClaw-based AI employees that worked nonstop and could assign tasks to humans
  • The team had to modify OpenClaw's architecture to add role-based permissions after it initially shared all company data with anyone who asked
  • Wu estimates 60-70% of work can now be handled by AI employees, calling it "a no-going-back experience"

The Signal

This is the first real operational report from inside a company running autonomous AI employees, and the details matter. Kuse modified OpenClaw because the original design treated enterprise data like a personal assistant would treat your calendar. No permissions layer. No role awareness. Just an AI that would cheerfully hand your financial data to anyone who asked.

The fix tells you where agent infrastructure is heading. They rebuilt OpenClaw to understand organizational hierarchy, who reports to whom, what each role can access. The AI employees now know the difference between a junior developer and a CFO. That's table stakes for Web4, but apparently OpenClaw shipped without it.

The human-only Slack channel is the detail that makes this real. When your AI colleagues work 24/7 and can delegate to you, you need somewhere to catch your breath. That's not a bug in the agent economy. That's what happens when automation doesn't sleep and humans still do. The boundary between human and agent labor just became a chat channel with restricted membership.

Wu's 60-70% automation estimate tracks with what we're seeing elsewhere. Not full replacement. Augmentation at scale. The work that gets automated first is the work that already felt mechanical. What's left is the creative stuff, the judgment calls, the problems that don't fit templates.

The Implication

If you're building agent infrastructure, permissions and role awareness aren't features. They're the foundation. Personal AI assistants don't translate to enterprise without rearchitecting for access control. And if you're managing humans alongside AI employees, you'll need spaces where humans can coordinate without agents listening. The org chart just got more complicated.

Source: Business Insider Tech