The AI-Generated Terraform Plan Looked Fine: Reviewing AI Infrastructure Code

The dangerous AI failure in infrastructure work is not the obviously broken output. It is the plausible one. A generated Terraform change that names sensible resources, reaches for the right provider blocks, and produces a plan that scrolls past without anything screaming is the change most likely to reach production carrying an assumption nobody checked — precisely because it never tripped the instinct that makes a reviewer slow down.

This is a five-minute talk with one job: name that pattern out loud and hand the room the review habits that catch it. The argument is not that AI writes bad infrastructure code. It is that AI writes infrastructure code plausible enough to stop the reviewer asking the right questions — and that the plan output, the one artifact that actually states what is about to change, is the thing we skim hardest when it looks calm.