Agentic AI is a core innovation in my research, enabling autonomous, closed-loop catalyst discovery by integrating simulation, machine learning, and decision-making into a unified framework. Building on our prior work in DFT, kinetic modeling, AI, and generative inverse design, we develop multi-agent systems that can plan, execute, evaluate, and refine catalyst design strategies under uncertainty. In this approach, a large language model (LLM) planner coordinates specialized agents, including predictive models (e.g., GNNs), diffusion-based generative models, and active learning modules, to iteratively explore the catalyst–reactor design space. These agents leverage physics-informed datasets and multiscale simulations to propose new materials, trigger targeted computations, and update models based on performance feedback. By embedding domain knowledge, mechanistic constraints, and experimental validation into the learning loop, the agentic AI system moves beyond static prediction toward adaptive, self-improving discovery. This framework significantly accelerates innovation, reduces computational and experimental cost, and enables scalable, interpretable, and autonomous design of catalytic systems.