In this talk, I will discuss what remains for physicists to do in the age of AI, with a focus on how machine learning is transforming scientific discovery and engineering in optics and photonics. I will highlight both opportunities and limitations of current AI approaches, and how physics-based reasoning, interpretability, and uncertainty quantification remain essential. I will also present a few concrete examples where combining physical models with data-driven methods enables more robust inference, design, and control in complex systems. Finally, I will outline how physicists can contribute by shaping AI tools that are reliable, explainable, and grounded in fundamental principles.