0:46
In this excerpt, Hamel Husain argues that AI evaluation metrics are unsuitable for inclusion in employee performance reviews because they are unstable and tend to be goal‑seeking—shifting as objectives evolve. He recommends that performance assessments for roles such as customer support agents be anchored to concrete product or business metrics (e.g., resolution rate, average handle time) that demonstrably improve over time. Husain warns that if all eval metrics consistently pass, they lose discriminative power and become effectively useless.
