Paper Type

Short

Paper Number

PACIS2026-1557

Description

Agentic generative AI (GenAI) is shifting from passive advice to active, reviewable action in policy-bounded organizational service work, creating a control challenge: users must delegate, monitor, intervene, and recover from AI-driven uncertainty and errors. This study theorizes calibrated reliance as emerging from an integrated agentic GenAI control architecture comprising interaction autonomy mode, uncertainty transparency, and override friction. Grounded in cybernetic control theory and productive delegation, the model predicts a three-way configurational effect on reliance calibration. It further theorizes two parallel but tensioned downstream pathways: effort burden as a resource-depleting oversight cost and epistemic growth as a capability-building learning gain, which jointly shape perceived user GenAI resilience. A 2×2×2 between-subjects online experiment in a customer-support ticket triage task will test the model. The study refines theories of control, delegation, and digital resilience while offering design guidance for safer and more sustainable human–AI collaboration.

Comments

02-FutureofWork

Share

COinS
 
Jul 5th, 12:00 AM

Calibrating Human Control in Agentic AI: How Autonomy, Uncertainty Transparency, and Override Friction Shape User GenAI Resilience

Agentic generative AI (GenAI) is shifting from passive advice to active, reviewable action in policy-bounded organizational service work, creating a control challenge: users must delegate, monitor, intervene, and recover from AI-driven uncertainty and errors. This study theorizes calibrated reliance as emerging from an integrated agentic GenAI control architecture comprising interaction autonomy mode, uncertainty transparency, and override friction. Grounded in cybernetic control theory and productive delegation, the model predicts a three-way configurational effect on reliance calibration. It further theorizes two parallel but tensioned downstream pathways: effort burden as a resource-depleting oversight cost and epistemic growth as a capability-building learning gain, which jointly shape perceived user GenAI resilience. A 2×2×2 between-subjects online experiment in a customer-support ticket triage task will test the model. The study refines theories of control, delegation, and digital resilience while offering design guidance for safer and more sustainable human–AI collaboration.