Eventually someone will forget. Of course, it happened.
The team copy-pasted Kubernetes configs across dev/staging/prod folders. Changed replica count in dev. Updated staging. Forgot prod.
Prod went down.
The problem: Manual copy-paste doesn’t scale. Human error is inevitable.
The solution: Automation. But there are three different tools, and the choice isn’t obvious.
Short version:
- Manual copying = will break eventually
- Helm = powerful but complex for simple tasks
- Kustomize = simple but not for everything
- Hybrid = what mature teams use
Result after switching: deployment errors -80%, deployment time cut in half.
⚙️ Technical details:
Three approaches:
Raw manifests = copy-paste YAML files per environment. Simple at first, nightmare at scale. Every change multiplies by number of environments.
Helm = package manager (like npm for K8s). You have base values.yaml (or defaults in the chart itself), then override per environment:
helm install app -f base-values.yaml -f dev-values.yaml --set image.tag=v1.2.3
Powerful templating with Go templates. Great for packaged apps with many config knobs.
Kustomize = layering system. Base manifests + patches per environment. No templating language - pure YAML overlays. What you see is what deploys.
Common mistakes:
- Raw manifests in prod (copy-paste inevitably drifts)
- Helm for everything (template syntax pain for trivial changes)
- Kustomize for third-party apps (300-line patches = hell)
- Not knowing you can mix them
What actually works:
Raw manifests when:
- Proof of concept, single environment
- That’s it. Seriously, don’t do this in production.
Helm when:
- Third-party software (Prometheus, databases, ingress controllers)
- Complex apps with many deployment variations
- Sharing reusable packages across teams
- Need rollback/upgrade lifecycle management
Kustomize when:
- Your own microservices across environments
- Simple environment patches (replicas, limits, domains)
- GitOps workflows without template debugging
- Patching third-party manifests without forking
The hybrid:
Helm for dependencies (with env-specific values files), Kustomize for your apps. This is what mature teams run. ArgoCD supports both natively for a reason.
Real example from an audit: moved from copy-pasted manifests to Helm (infra) + Kustomize (apps). Deployment errors dropped 80%, pipeline time cut by half.
Your setup? All-in on Helm, Kustomize only, or mix of both? Or something else?