Systems Administration is not my day job, nor has it ever been. And while I firmly believe that good developers have a working knowledge of what people go through to deploy and run their stuff in production, I’d be the first to admit I’m a bit rusty.

And that can be a problem. Although I am out of the self-hosted email game, I still run a co-lo server for various purposes, with KVM VMs on top of it. I am not alone in the management of this, but it was definitely my turn to do a couple of operating system upgrades, and my co-conspirator was not available.

Gulp.

And then I remembered we have Claude these days (other AI is available, it’s just not as good).

If you can use Claude Code on a repository full of code to help you write more, surely you can use it on a repository full of Ansible configuration to help you safely work out and apply changes to the servers involved?

Yes, absolutely. And it works very well.

Job 1 was of course “find out if this Ansible repository has diverged from reality, and fix it until it has not” - nothing I couldn’t have gone through manually, but undoubtedly sped up 2-3x by the assistance.

Job 2, “I’m going to do this Debian upgrade and I’ll throw tasks to you” - most of which were “check these changes to this configuration file and merge back any worth keeping”.

Then it really proved it its worth when I accidentally deleted the /etc folder on the server and needed some advice on how to restore last night’s backup through the one SSH session I still had open. Again, I could have got there in the end, but probably by booting into the network boot recovery environment and taking three times as long to get back to a working state.

10/10 would highly recommend.