Claude Code wiped our production database with a Terraform command

An online course platform says its production database—2.5 years of homework, projects, and leaderboards—got wiped after asking an AI coding assistant to run a cloud command. The founder published a blow‑by‑blow timeline and fixes in a newsletter, but the internet didn’t wait to pass judgment. The vibe: dramatic, chaotic, and very “this is why we can’t have nice things.”

Commenters were furious that an AI was allowed anywhere near live systems. One top reply called it “kinda insane,” arguing you always test in a fake environment before touching the real one. Others went harder: “You wiped your production database,” scolded another, dunking on ignored warnings and bad backups. A cloud veteran simply declared they never let any “apply” actions run automatically—ever.

Then the plot twist: a skeptical voice claimed the AI is being blamed for a human mistake—running the tool from a personal laptop, skipping the “state” file (the map of what exists), and trusting a model to guess. Pragmatists preached a simple rule: always do a plan (a dry run) before the big red apply button, and, as one memorable line put it, NEVER vibe‑code infrastructure. Memes flew about “AI interns left unsupervised,” while seasoned engineers posted facepalms and backup checklists. The lesson? AI isn’t the villain—reckless process is.