Sam Rodriques discusses Kosmos, an AI agent that can condense months of scientific analysis into hours by synthesizing vast amounts of data and generating code autonomously. While such tools accelerate hypothesis generation and data interpretation, they don’t replace experimentation or solve deep biological unknowns. AI’s real value lies in augmenting researchers, not replacing them—especially in parsing literature and optimizing study designs. However, major challenges remain: clinical translation is slow, funding is scarce, and serendipity still plays a crucial role in discovery. Overhyped concepts like virtual cells lack substance compared to tangible advances such as generative protein design. True progress requires clear benchmarks and integration with wet-lab validation. Despite the excitement, AI cannot bypass the fundamental need for biological insight and long-term trials.