Recent advances in ophthalmic AI have improved benchmark performance, yet clinical trust remains limited. We argue that progress should move beyond data and model scaling toward trustworthy, skill-efficient systems that integrate multimodal evidence, external knowledge, and uncertainty-aware reasoning. Ophthalmology provides a strong testbed for agentic AI, but safe clinical translation will require rigorous validation, workflow integration, and evaluation frameworks aligned with real-world decision making.
Jin et al. (2026) studied this question.