Over the past two decades, machine learning has transformed fluid mechanics and aerospace research by enabling the representation of nonlinear, high-dimensional relationships, data-driven turbulence models, accelerated numerical algorithms, and computational design. Yet these systems typically operate within narrowly prescribed tasks, leaving scientists to connect prior knowledge, formulate hypotheses, implement methods, construct computational tests, and interpret evidence.
Large language model agents, connected to scientific literature, data, coding environments, numerical solvers, and verification and validation tools, offer a broader capability. They can translate formulations into executable implementations, run and diagnose simulations, compare explanations, and design tests that can falsify them. Drawing on work in data-driven modeling, AI-assisted scientific discovery, and end-to-end CFD execution, this seminar examines how such agents can shorten the path from an idea to computational evidence and help produce durable scientific contributions—interpretable equations, scaling laws, algorithms, validated models, and new physical understanding. The goal is trustworthy agent-enabled research that preserves scientific accountability: agents provide breadth, speed, integration, and persistence, while scientists define consequential problems, impose physical constraints, establish standards of evidence, and decide what constitutes understanding.