Use of AI
Engineers are given a degree of freedom to integrate AI into their workflow as they see fit, however some guardrails do exist.
NOTE
As this is a space that is emerging and moving relatively quickly, some details may be short-lived.
Our stance
Our approach to AI adoption as an engineering team can be summarised as cautious and pragmatic adopters.
- We want to take advantage of the vast technological progression that is happening in this field
- If we don’t adopt AI we risk falling behind – as a team or as individuals
However…
- Being on the bleeding edge hurts (the clue is in the name)
- It takes time to adapt to new ways of working
- We don’t yet know the true cost
So we will…
- Take an iterative approach
- Only take advantage of new tools as they mature to avoid unnecessary risks and disruption
- Keep a tight control on costs
Providers
We have officially adopted Cursor as our primary engineering tool, and Gemini in the broader context of the company.
| Purpose | Example use cases | Context | Plan | |
|---|---|---|---|---|
| Cursor | Engineering lifecycle work | Generating code, generating tests, answering questions about codebases, investigating bugs, debugging, technical QA | Cursor app (agents), Cursor IDE, open coding harnesses (e.g. OpenCode) | Team |
| Gemini | General LLM work | Anything that doesn’t need the context of a codebase or the engineer mindset, e.g. non-engineering questions, workflow suggestions, competitor analysis. | Gemini app (desktop or web) | Pro |
WARNING
⚠️ Use of new AI tools and service providers is subject to sign off from our legal department. Before adopting a new tool, speak to the Head of Engineering and/or the Head of Legal. The initial due-diligence is normally very quick.
Accountability
Remember that when you use AI to do work on your behalf, you are still responsible for the contributions as if you made them yourself.
"I don't know, my AI did it"
…is not a satisfactory answer to…
"Why is this here?"
Of course, one of the main benefits of using AI is to speed things up, generating code in relation to a prompt or a task description, and as a result, engineers will have less line-by-line involvement with the code. However, the direction and architectural choices should always be reviewed and validated by human eyes.