Plug-in devs: how are you thinking about AI agents inside your build tooling?

I’ve been building my first iPlug2 projects recently and mostly lurk here reading how others structure their repos. Lately the news about California suing OpenAI over rogue AI agent incidents has me thinking about the tooling side of things. I’ve been experimenting with letting an AI agent run automated code changes on my plug-in source — mostly refactoring parameter handling and cleaning up UI layout code — and it’s convenient, but I keep worrying about an agent touching something in the audio processing path and me not noticing until it’s already committed.

I don’t have a solid process for this yet. Right now I manually inspect diffs, but on bigger changes it’s easy to skim. Curious what people here do: do you restrict agents to test files or UI code, use separate branches, or skip them entirely for sensitive DSP? Any workflow that’s worked well for you would be helpful to hear about.

Update after reading a bit more - California sues OpenAI over rogue AI agent incidents came up while I was searching for how others handle agent permissions. I’m still not sure restricting agents to UI code would be enough, but I’m leaning toward separate branches for anything near DSP.