Agentic AI, Biology, and What Remains Human (dvitsios.org)

The concept of Agentic AI is transforming workflows in fields like biology and pharmaceuticals by automating the entire process of planning, coding, testing, and iterating. This shift enables faster exploration of possibilities but raises crucial questions about human oversight: rather than merely supervising outputs, humans must now focus on steering the exploration towards relevant questions and safe outcomes. The rise of Agentic AI challenges the established notion of productivity, emphasizing the need to refine how tests are designed, as effective testing now encapsulates compressed human judgment, vital for contextual correctness. As these AI systems evolve, human roles are shifting from direct supervision to strategic steering, where understanding the implications of results and decision-making becomes paramount. In scientific domains, nuances such as data retrieval errors can lead to significant misinterpretations, making it essential for humans to guide the AI toward meaningful inquiries and pathways. This newfound reliance on human judgment stresses the importance of resilience and adaptability in navigating a fast-changing landscape where the AI's capabilities will continue to evolve, potentially demanding a re-evaluation of how expertise and productivity are defined in the era of AI.