AIS Insights AI Podcast

https://ais-insights.ai/insights/851

Paper Number

ICIS2026-1345

Paper Type

Completed

Abstract

Agentic systems are beginning to automate knowledge work by planning, invoking tools, and iterating across multi-step tasks, yet little is known about how different autonomy levels shape human agency and performance. Drawing on the ironies-of-automation tradition, we examine two paradigms of human–agent collaboration: orchestration, in which users compose and supervise agentic workflows, and delegation, in which autonomous agents pursue user-defined goals. Using a design science research approach, we derive design requirements from the literature and interviews with knowledge workers, instantiate them in two software artifacts, and evaluate both in a within-subjects experiment on a voice-of-customer analysis task. We find a control–efficiency trade-off: orchestration supports closer alignment between perceived and objective performance at the cost of higher mental workload, whereas delegation enables higher task correctness but makes performance harder to assess. These findings respecify the ironies of automation for agentic systems and offer design knowledge for organizations.

Comments

03_Future_of_work

Share

COinS
 
Dec 14th, 12:00 AM

Orchestration or Delegation? Balancing Human Agency and AI Autonomy in Agentic Systems

Agentic systems are beginning to automate knowledge work by planning, invoking tools, and iterating across multi-step tasks, yet little is known about how different autonomy levels shape human agency and performance. Drawing on the ironies-of-automation tradition, we examine two paradigms of human–agent collaboration: orchestration, in which users compose and supervise agentic workflows, and delegation, in which autonomous agents pursue user-defined goals. Using a design science research approach, we derive design requirements from the literature and interviews with knowledge workers, instantiate them in two software artifacts, and evaluate both in a within-subjects experiment on a voice-of-customer analysis task. We find a control–efficiency trade-off: orchestration supports closer alignment between perceived and objective performance at the cost of higher mental workload, whereas delegation enables higher task correctness but makes performance harder to assess. These findings respecify the ironies of automation for agentic systems and offer design knowledge for organizations.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.