I Let AI Run My Sysadmin Tasks… Here’s What Happened

A sysadmin shares a multi-week experiment using Claude Code on a large personal Ansible automation project (300+ files, 13,000+ lines). Despite initial skepticism, Claude Code proved surprisingly capable at understanding complex project architecture, identifying issues like missing DNF5 collision checks, and making targeted code changes. Key takeaways: AI is useful as a productivity multiplier but requires human review of all changes, should not run unattended, and must not be given access to proprietary or confidential data. The author compares AI to a junior developer whose work always needs auditing.