Emerging Practices for AI in Research Software Engineering: A Spectrum from EVERSE RSQKit

Abstract

AI may democratise code-based innovation in ways not seen since the creation of the modern spreadsheet. Research software is part of this new shift. But tools that broaden automation beyond specialist programmers are evolving quickly, and sensible ways to use them are still emerging. This is especially important in the context of guiding and supporting researchers who code, where the experience of RSEs becomes more important as AI use grows. This is particularly true given security considerations around AI-generated code and the clear need for review and mitigation strategies. RSEs therefore need a practical model for deciding how to use AI across a spectrum of AI assistance while preserving quality. Drawing on experience from EVERSE [1] and RSQKit [2], its Research Software Quality Toolkit, this talk presents a spectrum of AI assistance: from direct prompting and tool integration through coding assistants to more agentic approaches. It examines where these approaches help in practice, including drafting, documentation, testing, support, and routine development work. It also considers how AI use will affect quality, reproducibility, transparency, and appropriate credit, and where judgement, review, and human control remain essential. The talk closes with practical suggestions for RSEs, projects, infrastructures, and institutions seeking to adopt AI. The EVERSE project is a multi-organisation EU Funded collaboration, which aims to create a framework for research software and code excellence, collaboratively designed and championed by the research communities. [1] https://everse.software/[2] https://everse.software/RSQKit/