fig1

Figure 1. Framework of closed-loop discovery driven by AI Research Agents. Unlike conversational demo agents, real research agents act as workflow orchestrators at the center of the materials discovery cycle, seamlessly connecting literature databases, theoretical constraints, experimental characterization, and structured feedback. The right panel illustrates the paradigm shift toward “Delivery Agents”, where models delegate tasks to tools to execute physical experiments. Bottom panel outlines the four critical requirements - Reliability, Verifiability, Integration, and ROI - necessary for transitioning agents from conceptual demonstrations to deployable research infrastructure. AI: Artificial intelligence; ROI: return on investment.