Sarah discusses the importance of running experiments to evaluate the impact of new features on model performance without needing to access data directly. She emphasizes the necessity of understanding potential regressions and how these insights can benefit numerous customers utilizing auto ML.
