The author shares insights from an experiment assessing how candidates perform in coding interviews when given access to AI tools like Claude or ChatGPT. The findings reveal a clear divide: strong candidates benefit from AI, while weak candidates do not. Those who already demonstrate solid problem-solving skills leverage AI effectively, refining their approach and accelerating their progress. They formulate specific prompts, analyze the AI's responses, and adjust their strategies accordingly, leading them to solutions faster. In contrast, weaker candidates struggle to use AI productively. They tend to rely on vague prompts and fail to comprehend the AI's feedback. For instance, they might ask, "What's wrong with this program?" without fully grasping the response. This lack of clarity and focus results in little to no improvement in their performance. The author had hoped that AI would provide a safety net for these candidates, but it instead magnifies their existing ineffective habits. Looking ahead, the author speculates that as AI technology advances, it may eventually guide users away from poor problem-solving practices. Current AI capabilities, however, still require clear and logical thinking from users to be effective. The article emphasizes that without foundational skills, AI tools won't significantly enhance performance in high-pressure situations like coding interviews.