AI: Artificial Intelligence
LLM: Large-Language Model
CLI: Command Line Interface
IDE: Integrated Development Environment
ARC: Advanced Research Computing
This guide is a collection of advice and tips based on the experience of ARC staff. ARC’s position is that we encourage curiosity, exploration and experimentation, including with AI-assisted coding tools. We hope this guide can act as a gateway for anyone interested in this area.
Disclaimer:
These are some general guidelines and pointers around AI-assisted coding tools, based on the individual experiences of ARC staff. It is intended to be a collaborative document and we invite any ARC staff interested in this topic to contribute. Feel free to open a pull request to improve the content. ARC staff should use the #community-ai-assisted-coding channel to discuss anything they are unsure about.
Read the ISD AI Practices in Software Development (only available to UCL staff) for a comprehensive overview of AI/LLMs and how to use them responsibly. The European Commission’s Living guidelines on the responsible use of generative AI in research is also worth a read.
The RSE AI position statement discusses the role of AI for RSE work specifically (our director is a signatory).
There are a lot of tools out there. At the time of writing (December 2025), AI agents are all the rage and this is what most of the tools listed below rely on. AI agents go beyond the simple chatbot interfaces such as ChatGPT, and provide more automation by integrating with your other tools, like your IDE, to carry out tasks for you (like editing code). You prompt them in a way similar to chatbots, but you can also provide them with more specific guidelines for your particular project (sort of like a contributing guide, but for agents instead of humans). Agents.md is an attempt at standardising this across agent providers, though not every provider supports it yet (Claude is a notable example, though they provide a workaround).
The actual Large Language Models that power AI tools. Too many to list and keep track of, but some popular choices are:
Choosing a model for your task often comes down to personal preference and experimentation
These let you run and interact with AI agents straight from your terminal and let them carry out tasks you would do from a terminal. And yes, you can absolutely nuke your filesystem with them if you’re not careful.