People-First, Collaborative AI: How we use Code Assistants and AI Tools in our Business.

Coding with AI assistance. Image generated using AI.

Like every business, we find ourselves using AI in the day-to-day workflows of business. Having seen technologies come and go over the years, and with the incredibly fast-moving nature of AI, we are adopting it optimistically, but cautiously, into our business

A particularly important area to consider is writing code, which is one of the core operations of the business! Whilst a lot of noise surrounds AI and vibe coding, handing off coding tasks to AI agents and automation, this is not our approach.

Instead, we use AI very much as an assistant, or a collaborative partner, to the improve the coding process, and help debug issues. AI is fantastic and quickly identifying where issues may occur, scouring obscure and often overly-complicated documentation, and delving into open-source code to surface important details, considerations, or issues,

We never use AI for copy/paste code into production codebases. We treat AI's suggestions as just that - a suggestion, with every line of code examined and refined by a human developer before it is commited to the repository. This approch helps us to safeguard quality, security, and ensure that best-practice standards are upheld across the codebases and architecure of the systems we build and maintain.

We also leverage AI for routine, time‑consuming chores such as generating tests and specifications, helping to refactor code blocks. This allows us to optimise our time, offloading necessary by mechanical tasks, so we can focus on the activities and outcomes valued by our customers.

In practical terms, we make day-to-day use of Github's Co-Pilot service, which is our primary AI code tool. We also use privacy preserving AI such as Proton's Lumo and DuckDuckGo's duck.ai, and Canva’s AI for generating some images for our marketing content (such as the image above). We are also testing AI locally using a combination of Ollama and AnythingLLM running in local Docker containers.

As well as inline assistance during the coding process, we've also recently started using Co-Pilot to perform Pull Reviews, which has proven very helpful in summarising changes to provide a more complete documentation and audit trail, and for spotting minor errors, typos and erroneous whitespace (Co-Pilot's "nitpicks") that may have made it into the PR.

There is no denying that AI is an incredibly useful tool in the kit available to human developers - and it will only get better with time - but it is not a replacement. AI augments people's talent, accelerates troubleshooting, and helps us to deliver solutions and value faster. It's important, though, that the final output - the code that makes it to production - is always fully reviewed and authored by experienced human developers, ensuring that the software we ship remains reliable, secure, and aligned with our customers' business goals.