diff --git a/hosts/k8s/README.md b/hosts/k8s/README.md
index 54d7c0c7..ca83f371 100644
--- a/hosts/k8s/README.md
+++ b/hosts/k8s/README.md
@@ -23,5 +23,9 @@ I prefer to use [k3s] as the Kubernetes distribution, because it's lightweight,
    3. `k3s-test-1-worker-2`
    4. `k3s-test-1-worker-3`
 
+## Kubernetes Resources
+
+Kubernetes resouces are deployed and managed separately through [nix-config/pulumi/k8s/](../../pulumi/k8s/).
+
 [k3s]: https://github.com/k3s-io/k3s/
 [what-have-k3s-removed-from-upstream-kubernetes]: https://github.com/k3s-io/k3s/?tab=readme-ov-file#what-have-you-removed-from-upstream-kubernetes
diff --git a/pulumi/README.md b/pulumi/README.md
new file mode 100644
index 00000000..cc8e8b48
--- /dev/null
+++ b/pulumi/README.md
@@ -0,0 +1,4 @@
+# Pulumi - Infrastructure as Code
+
+TODO
+
diff --git a/pulumi/k8s/README.md b/pulumi/k8s/README.md
new file mode 100644
index 00000000..f20b6cf1
--- /dev/null
+++ b/pulumi/k8s/README.md
@@ -0,0 +1,20 @@
+# Kubernetes
+
+
+## Why Pulumi for Kubernetes?
+
+1. Deploying Helm charts declaratively.
+   - Helm CLI supports only imperative commands, you need to add a repository, update the repository, 
+     and install the chart with a single command,
+     it's really hard to manage the lifecycle of the Helm chart in this way.
+    - Pulumi can deploy Helm charts declaratively, you can manage the lifecycle of the Helm chart easily.
+1. Deal with secrets in a secure way.
+1. Deploying Kubernetes resources in a unified way, instead of running a bunch of commands like `kubectl apply`, `helm install`, `kustomize`, etc.
+
+## Why not ArgoCD or FluxCD?
+
+ArgoCD & FluxCD support only Kubernetes, and it's too heavy for my use case.
+
+Pulumi supports not only Kubernetes but also other cloud providers like Proxmox, Libvirt, AWS, Azure, GCP, etc.
+It's a unified way to manage the lifecycle of all my infrastructure resources.
+