How I use AI in development

I manage how AI tools are adopted in the repository and in the teams’ development work. I use and evaluate Claude, Cursor and other assistants to analyse code, prepare changes and reduce repetitive work. I define usage rules and maintain the repository context available to them.

I pay attention to recurring mistakes: losing context between modules, overlooking an existing implementation or suggesting changes outside the task. I investigate what is missing from the instructions, documentation or shared tooling before trying again.

What I look at:

1. The task and its context. I check what the repository already provides and which modules the change should touch. An agent proposing duplicate functionality is a useful sign that it has missed part of that context.
2. The recurring problem. If the same mistake keeps coming back, I check whether an instruction is unclear, documentation is missing or a shared tool needs changing.
3. The resulting code. I review the diff, run tests and check the integration with the project. I keep changes I can explain and maintain; a plausible answer alone does not tell me whether the software works.