This puts a box around a phenomenon I’ve been seeing. While there are real challenges with AI slop, quality AI assisted work is devalued.

Our experiment showed no quality difference in AI-assisted work… This disconnect reveals the path forward—shift from evaluating how work is done to what is accomplished. Replace subjective competence ratings with objective metrics such as cycle time, accuracy, defect rates.

https://hbr.org/2025/08/research-the-hidden-penalty-of-using-ai-at-work