Agentic AI Moves from Tool to Scientific Collaborator

The most significant shift in technology research is not a new material or algorithm but a change in who—or what—conducts the research. Agentic AI, systems capable of autonomous decision-making and action, is moving beyond the chatbot paradigm into the laboratory and the research seminar. In late 2025 and through 2026, multiple agentic AIs have begun forming specialized teams within computers, autonomously carrying out complex research tasks while discussing and issuing instructions to one another . This is not merely an extension of generative AI’s text-generation capabilities; it represents a discontinuous evolution in how scientific inquiry itself is structured. RIKEN, Japan’s flagship research institute, has launched the RIKEN AI for X-disciplinary Initiative in Science (RIKEN-AXIS) explicitly to harness this shift, integrating AI with cutting-edge measurement facilities and the Fugaku supercomputer to build an open scientific research platform .

The practical applications are already emerging in high-stakes domains. Harvard Medical School’s AURORA project, funded by Cancer Grand Challenges with nearly $7 million, is developing AI “co-scientists” that work alongside human researchers through the full cycle of discovery—analyzing vast data pools, generating testable hypotheses, suggesting experiments, and refining reasoning as results come in . Unlike chatbots, these systems access scientific literature, databases, biological AI tools like AlphaFold, and real-time experimental findings, prioritizing hypotheses based on time, budget, or other constraints important to human investigators . Three international research teams will evaluate the system in studies of CAR T-cell therapies, cancer initiation, and tumor-immune interaction, with humans remaining responsible for all scientific judgment and decision-making .

The deeper question this research raises is philosophical as much as technical. If AI can generate hypotheses that humans might miss and design experiments more efficiently than human effort alone, what becomes of the researcher’s role? RIKEN’s president frames the answer in terms of co-creation: humans must refine the wisdom needed to make sophisticated use of AI’s power as it evolves . The goal is not to replace scientific judgment but to expand the space of exploration—to use AI to enrich what RIKEN calls “the joy and art of discovery” . Whether this vision materializes depends on whether these systems can be trusted to fail transparently, learn from experimental feedback, and integrate with the messy, iterative reality of laboratory science.