Granting AI agents the autonomy to intervene and relinquish control during high-pressure workflows does not reliably reduce cognitive load. A study conducted by Lingnan University and published in the International Journal of Human–Computer Studies reveals that autonomous task handoffs undermine human situational awareness, trigger higher mental fatigue, and spark direct authority conflicts between operators and machines.
Led by Professor Jie (Jay) Xu, director of Lingnan University’s Cognitive Science Research Center, researchers tested 92 participants on NASA's multi-attribute task battery (MATB) in flight simulation environments. While shared authority allocation—where algorithms detect workload spikes and proactively seize control—stabilized tracking accuracy during acute surges, the model crumbled the moment pressure subsided. When the system abruptly returned control, operators suffered context detachment, struggled to reorient within active workflows, and recorded significantly higher late-stage fatigue compared to human-led authority allocation.
Autonomous intervention does not eliminate cognitive effort; it merely shifts the burden from active execution to exhausting, continuous surveillance. As Professor Xu's findings demonstrate, opaque agent decision logic forces operators to spend vital cognitive bandwidth tracking algorithmic behavior, decoding intent, and guessing the next handoff. For enterprise leaders, uncontrolled agentic autonomy introduces hidden operational overhead: scalable human-AI architecture requires rigorous transparency and deliberate, human-governed authority handoffs rather than unprompted machine takeovers.