The Prevalence of Smells in AI‑Generated Kubernetes Manifests

As Kubernetes adoption surges, so does the use of AI to generate deployment manifests—promising speed and simplicity. But is convenience worth the risk? In our latest study, we analyzed 98 real-world Kubernetes manifests generated by ChatGPT, uncovering a sobering truth: nearly half contain critical configuration smells—like unbounded resource limits, containers running as root, and dangling services. Our findings reveal that AI-generated code isn’t inherently flawed, but it does require rigorous safeguards. With security and operational risks on the rise—especially as developers grow more reliant on AI—we argue for an essential shift: static analysis must be woven into every CI/CD pipeline. Learn how tools like KubeLinter, enforced policies, and developer education can turn AI from a liability into a reliable ally. The future of cloud-native infrastructure isn’t just automated—it must be responsible.