doi.org
AI-Assisted Software Development Policy for UCL Advanced Research Computing
This policy covers the use of large language models (LLMs) by ARC staff for software, systems and platform engineering, data analytics and computational statistics, modelling and simulation, and process automation — briefly "AI for programming". It covers model and tool selection, collaborative behaviour in teams, collaborator/customer interactions, future research directions, and education and training content for teaching software and systems engineering.A comprehensive review of AI-assisted software development, from tab-completion to multi-agent harnesses such as Codex and Claude Code, is beyond the scope of this policy. This policy does not directly address the broader questions surrounding LLMs: the provenance and copyright status, environmental impact, long-term labour market impact, and social and cultural implications, among others. Those are legitimate and live debates, and colleagues who have concerns should raise them within the appropriate ARC fora, their team and with line managers. We encourage open discussion of these issues in a spirit of disagreeing well.Creating a consensus policy is hard. This document is the result of extensive discussions within UCL's Advanced Research Computing Centre, and endeavours to set out principles and an overall position within which people holding diverse views on the topic can work. The authors acknowledge and are grateful for rich discussions and contributions from many people beyond those listed.
