Has Anyone Successfully Used Claude Code or Codex for Real DevOps Tasks?

I've tested out Claude Code, Codex, and the Gemini CLI, but I ultimately prefer Claude. Codex usually gives me unreadable shell command edits, but Claude's diffs are much clearer. I regularly use it in both staging and production, but I always review everything manually. I never just let it do things without understanding the proposed changes first.

I love using Claude for Terraform because it works surprisingly well! However, I usually find that many models can be outdated, which limits their effectiveness. But it does a better job now at helping identify the root causes of issues.

We've been leveraging it to write OPA policy controls and tidy up our setup. However, when it comes to creating infrastructure, it can still hallucinate quite a bit. Often, it suggests resources that don't even exist. So, I think it should be used cautiously and thoroughly reviewed before implementing any changes.

I'm using tools like Cursor and Opencode alongside Claude Code, mostly to navigate terraform codebases I’m not familiar with. It’s super helpful for identifying errors based on context and quickly pointing out the root cause of issues. It really acts like a smarter search engine specifically for code, which makes diagnosing problems much easier.

I've been in DevOps and systems engineering for over 25 years, and I've used Claude Code and Codex daily for about a year and a half now. It's been a game changer for my efficiency! For everything from last-minute refactoring in hundreds of files to figuring out what new scripts do, it's helped me do all of those tasks in around 10 minutes. I even vibe-coded a whole data analytics platform in GCP with it, and the documentation is top-notch. Couldn't ask for anything better!