Authors: James Hetherington (University College London), Neil Chue Hong (EPCC at the University of Edinburgh), Daniel Katz (University of Illinois Urbana-Champaign), Robert Haines (University of Manchester), Sandra Gesing (US Association of Research Software Engineers), Charles Catlett (Argonne National Laboratory)
Abstract: Software engineering is changing rapidly as AI-assisted programming becomes accepted. Tools and approaches have matured rapidly from “vibe-coding” to using complex and highly productive agentic harnesses. In this session, we will discuss the implications for the scientific programming community, with a particular focus on HPC software. This includes systemware, middleware, and tools used to operate, manage, and maintain HPC systems, as well as parallel simulation and modelling software used in HPC settings. Topics include trustworthiness and validation for AI-generated software, career futures and professional implications for systems engineers and research software engineers, security implications, and emerging current "best performance practices."
Long Description: Goal & timeliness of the topic: Scientific engineering communities are attempting to quickly adjusting culturally to an enormous change in our working practices and identity: from programmers to agent-shepherds. Is this wise? Is this safe? Is this sustainable for us as a community, culture, and profession? How do we best adjust our norms and professional cultures? While much has been written about the general use of AI within software engineering, little is specifically focused on the needs of HPC codes and software ecosystems. This includes the relative underrepresentation of HPC programming languages and approaches in general AI tools, the increased challenges of ensuring scientific integrity on large-scale codes, and the dangers of losing HPC skills capacity in the workforce. The goals of this BoF are to advance ongoing discussions with additional community members' experiences, to begin to answer these questions, and to widen the discussion beyond its initial focus on research software engineers to include more members of the HPC community.
Community building: This BoOF seeks to bring together those reflecting on this, from enthusiasts to skeptics and everyone in between. The organising team includes those behind related workshops, including a Schmidt-Sciences funded international invitational workshop in Edinburgh, UK (https://www.researchsoft.org/events/rse-ai-workshop/), and discussion sessions at RSECon25UKRSE and USRSE'25 conferences, and a community position statement ( https://github.com/Academic-Data-Science-Alliance/rse-ai-position-statement/blob/main/RSE-AI-Final_Statement.md). The team also draws from the Trillion Parameter Consortium (TPC), an international community of ~1800 participants from more than 100 organizations. Focused on topics at the nexus of science, HPC, and AI, TPC’s technical working groups facilitate both best practices discussions and quarterly hackathons that support collaborative projects building AI capabilities and systems. Drawing participants from these and other communities, the The aim of this BoF is not to come to a consensus of the value of AI-assisted software engineering, but to share evidence and experience around its use to improve understanding, guidance and practice in HPC.
Relevance to the expected audience: software is a key element of HPC, developed and maintained in academia, laboratories, and industry, to operate, manage, maintain, and make productive use of HPC systems. Improvements in how it is developed and maintained will lead to improvements in the systems, their operation, and the work that is done using them. All members of the HPC community have a stake in this discussion.