What HAT performance means in practice

Human–autonomous team (HAT) performance is often reduced to throughput or error rate. Those matter, but they miss how people and agents share goals, attention, and recovery when the plan breaks.

In practice I look at three layers: outcome (did the joint system succeed?), process (was the division of labor understandable?), and experience (did the human stay calibrated — neither over-trusting nor fighting the agent?).

That framing travels well from lab studies into industrial environments where the “agent” might be a robot arm, a scheduler, or a driving stack.