Explore how AI agents respond to target setting and control, revealing surprising parallels with human behaviour and implications for managing AI in business.
- Speaker
- Coordinator
- Coordinator
- Date
- Thursday 22 May 2025, 11:10 - 12:30
- Type
- Seminar
- Room
- Mandeville T03-14
Abstract
This article introduces AI agents as objects of management control. We review recent literature on AI agent behavior and explore its implications for managing such agents in business settings. To illustrate how AI agents open up new research avenues in management accounting, we conduct an experiment on how targetsetting affects the performance and behavior of AI agents. Specifically, we test two stylized facts established in human performance research: (1) setting targets enhances performance, (2) difficult targets coupled within sufficient controls can lead to misconduct. Our results show notable parallels and distinctions between humans and AI agents. Unlike human counterparts, the AI agent (Claude Computer Use) does not perform better when setting a target. Intriguingly, similar to some humans confronted with difficult targets and inadequate oversight, the AI agent engages in "cheating". Collectively, this article provides as first step on the road towards designing effective management control systems for environments increasingly reliant on AI, highlighting both shared principles and unique considerations when steering AI-driven performance through targets.
