What if your research computing worked like a conversation? You describe the experiment; an AI agent picks the right machine, writes the code, submits the jobs, watches them run, fixes what breaks and reports back with results and provenance. In this talk, Paul Liu, the director of the NC State AI Hub for Science, will show how to build exactly that on infrastructure already available at NC State, without a big team or a big budget: one person, a frontier coding agent and a set of carefully designed guardrails. Liu will demo ASCEND (Autonomous Scientific Computing Engine for Novel Discovery), an AI-powered working system that spans three NC State resources: the NCShare GPU cluster, the Hazel HPC system and a departmental single-GPU workstation. The demo will include a single "front door" that probes live cluster state and uses AI to route each job to the right resource; agents that notice mid-conversation when they are on the wrong machine and suggest moving; one-command reproduction of published ML results; automated, policy-respecting retrieval of papers; and a knowledge base where the agent's lessons stay labeled as hypotheses until a real run validates them. The guiding principle: The agent reasons, but a controlled and auditable layer acts. Along the way, Liu will share the recipes and hard-won gotchas — multiplexing through Duo, shared-GPU etiquette, scheduler vs. no-scheduler boxes, onboarding new users from a single installer — so attendees can leave with a blueprint for standing up their own personal autonomous computing assistant on university, industry or even your own resources.

Liu's talk is part of the Department of Computer Science's AI series.