MIT has developed a computational framework enabling embodied AI agents to evolve vision systems over generations by adapting to environmental constraints and tasks. Agents start with simple photoreceptors and evolve features like compound eyes for navigation or camera-type eyes for object discrimination, driven by task-specific rewards and physical constraints. This method simulates evolutionary processes algorithmically, providing a 'scientific sandbox' for exploring how diverse vision systems emerge. It bridges scientific inquiry into evolutionary biology and engineering applications, potentially informing development of cameras and sensors tailored for robotics, drones, and wearables. The framework incorporates genetic encoding to mimic morphological, optical, and neural evolution and supports hypotheses about vision evolution otherwise inaccessible experimentally. Future expansions aim to integrate large language models for broader inquiry. This interdisciplinary effort merges AI, biology, and cognitive science under constraints resembling natural evolution.