Kitzy - How to actually use AI to manage your fleet
I made the case last month that without GitOps, there is no AI-accelerated device management. The AI needs a surface to work against, and a GUI isn’t one. That part stands. But I left out the next question: assume you have the repo. Now what? One thing before we get into it: the examples below are Fleet and Claude Code because those are the tools I use. The pattern is the same for any MDM with a config-as-code surface and any AI coding agent - your specifics will differ, but the workflow shape won’t. Translate as you read. If you point an AI coding agent at a vanilla MDM repo and prompt it to add a profile, you’ll get something back. It might even apply cleanly. It also might quietly overwrite an existing profile because it reused the identifier from the file it copied. It might reference an osquery column that doesn’t exist. It might use an app slug that fails at apply time because nobody told it to check the slug exists. It might wire a secret reference that expands to an empty strin