economics Paradigm Challenge

Companies actually make more money when their AI is 'risky' enough to scare employees—it forces everyone to stay on their toes and keep up.

March 25, 2026

Original Paper

Use it or Slowly Lose it: Expertise Atrophy with Organizational AI Usage 

James Siderius, Robert A. Shumsky, Alva Taylor

SSRN · 6398398

The Takeaway

We typically assume that skill atrophy is a net loss for firms, but this research shows that the potential for skill loss creates a self-motivating incentive for workers. Firms might actually prefer 'worse' AI or higher rates of skill loss because it forces humans to exert more effort to maintain their own expertise.

From the abstract

Problem definition: As organizations adopt generative AI, its use can improve productivity, but reliance can lead to atrophy of worker knowledge and skills. The challenge is how to incentivize human oversight and sustain long-run expertise. This is difficult because improving AI quality makes oversight and skill maintenance harder to motivate, as acceptable outcomes increasingly occur even when workers disengage. Methodology/results: Using a principal-agent framework, we study optimal incentive