Foundation models and AI agents in oncology drug discovery

Oncology drug discovery remains limited by high attrition, slow experimental iteration and weak translation from preclinical models to clinical benefit. This review examines how AI is restructuring that process through three connected layers: biological foundation models that reduce uncertainty in molecular and cellular systems; generative design methods that improve candidate quality and compress medicinal chemistry cycles; and autonomous discovery platforms that integrate reasoning, experimentation and feedback. We analyze targeted protein degradation, emerging clinical validation and evolving regulatory frameworks; and argue that future progress will depend on causal inference, context generalization, interpretability and regulatory-grade evidence generation rather than model scale alone.