← Back to Research
Google

Centaur Actors

Hybrid human-AI ensembles where one human directs many AI agents across diverse configurations throughout the day.

The term comes from chess, where centaur teams -- human plus computer -- consistently outperform both humans alone and computers alone. Google's Paradigms of Intelligence team extended this concept to the workplace: the future isn't AI replacing humans or humans supervising AI. It's fluid ensembles where the human moves between different agent configurations throughout the day.

In steadybase, this is exactly what happens. You open PipeScout, and Drew (your GTM agent) is working your pipeline. Switch to RollCall, and Scout is producing videos. Each app is a different centaur configuration with a different specialized agent. The 3-panel shell -- left nav, main content, agent chat -- is the cockpit of the centaur.

The key insight: the human doesn't need to understand how each agent works internally. They need to understand the output, make governance decisions, and redirect effort. The OS handles the rest.

Related Topics

See this in production

These research concepts power real agent deployments today. See how organizations are using the Agent Operating System.

Agents as a Service →