AI is getting good enough to generate Terraform, Kubernetes manifests, CI/CD workflows, and cloud configuration that looks reasonable at first glance. That is exactly what makes it risky.

This lightning talk walks through a simple but dangerous pattern: an AI-assisted infrastructure change that appears clean in a pull request, passes a quick human skim, and still carries bad assumptions into production. The issue usually is not that the AI wrote obviously broken code. The issue is that it produced something plausible enough that the reviewer stopped asking the right questions.

In five minutes, I’ll cover the review habits I now treat as non-negotiable for AI-assisted infrastructure work: checking intent against the actual plan, looking for hidden blast radius, validating provider behavior, separating generated convenience from operational ownership, and making sure policy guardrails catch what humans miss.

This is not an anti-AI talk. It is a practical reminder that infrastructure code is not just text. It changes real systems, real permissions, real network paths, and real production behavior. AI can help you move faster, but the Terraform plan still needs an adult in the room.