Agentic AI systems are beginning to transform scientific workflows by autonomously generating hypotheses, retrieving domain knowledge, orchestrating simulations, and interpreting results. In nuclear materials design—where multiscale physics, fragmented data, and long qualification timelines create a fundamental orchestration bottleneck—these systems offer a path toward accelerated discovery.

This talk presents an agent-based workflow for nuclear materials design that integrates simulation tools, machine learning models, and a structured knowledge base to enable iterative hypothesis–simulation–analysis loops. Using a case study on phase stability in U–Si systems, I show how agents can reproduce complex materials science analyses by making decisions about search spaces, model selection, and result interpretation—capabilities that begin to resemble those of a scientific collaborator rather than a traditional tool.

Access my slides here