video1.0<iframe src="https://www.loom.com/embed/7c7293efa0ba459ba2de243b0b5aacb2" frameborder="0" width="1366" height="1024" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>10241366Loomhttps://www.loom.com10241366https://cdn.loom.com/sessions/thumbnails/7c7293efa0ba459ba2de243b0b5aacb2-474ff0ff7dbe6da8.gif152.832In this Loom, I explain a cognitive load manager we built so an AI agent can learn to manage human workers sustainability, not just optimize tasks. We used a realistic multi agent environment with three worker agents, one AI manager agent, partial observability, deadlines, dependencies, and switching penalties. The manager assigns tasks, delays work, or gives breaks based on observed stress and task load, with an RL reward that tracks completions and burnout avoidance. We trained it in a 21.5D model and behaviors emerged like redistributing load from exhausted workers. No action is requested from you.