This paper investigates how physics students utilize generative AI (genAI) for computational modeling in open-ended assignments, revealing its significant impact on planning, implementing, and debugging code. It highlights both the efficiency gains and the risks to learning when students over-rely on genAI without critical verification.