Stanford computer science professor Michael Bernstein examines where artificial intelligence can be trusted to make decisions and where human judgment still has to carry the weight. Speaking in a Stanford Online webinar, he draws on his research into human-AI interaction to lay out frameworks for telling reliable AI use cases apart from ones that only look reliable. The talk covers how organizations evaluate AI adoption, what kinds of problems tend to fail silently when handed to automated systems, and how leaders can weigh the risk of over-trusting a model against the cost of under-using one. Rather than treating AI capability as a fixed line, Bernstein frames it as a moving boundary that shifts with the task, the data, and the amount of oversight built into the workflow, and he walks through concrete examples of augmentation succeeding and failing in practice.