This workshop addressed a gap: the most recent DOE reports on AI for science 1–3 predate recent advances in agentic AI—LLMbased systems that can plan, orchestrate, and execute multi-step workflows. With 100 participants from eight US DOE facilities and two allied facilities across accelerators, beamlines, nanoscience, genomics/biology, and computing infrastructure, this report identifies paths to deploy agentic AI at DOE user facilities while preserving safety, accountability, and rigor. Discussions also reflected DOE’s Genesis mission emphasis on end-to-end workflow acceleration through integrated data, computing, and AI capabilities.
Hellert et al. (Tue,) studied this question.