diff --git a/charts/operator/templates/deployment.yaml b/charts/operator/templates/deployment.yaml
index 74dc07d..2aa4769 100644
--- a/charts/operator/templates/deployment.yaml
+++ b/charts/operator/templates/deployment.yaml
@@ -13,6 +13,11 @@ spec:
       labels: {{- include "operator.selectorLabels" . | nindent 8 }}
       annotations:
         kubectl.kubernetes.io/default-container: manager
+        {{- if and .Values.metrics.enabled .Values.metrics.podScrapeAnnotations }}
+        prometheus.io/scrape: "true"
+        prometheus.io/port: {{ .Values.metrics.port | quote }}
+        prometheus.io/path: "/metrics"
+        {{- end }}
     spec:
       serviceAccountName: {{ include "operator.serviceAccountName" . }}
       priorityClassName: {{ .Values.priorityClassName }}
diff --git a/charts/operator/values.yaml b/charts/operator/values.yaml
index 48403ca..8cf0753 100644
--- a/charts/operator/values.yaml
+++ b/charts/operator/values.yaml
@@ -103,6 +103,14 @@ metrics:
     type: ClusterIP
   serviceMonitor:
     enabled: true
+  # Pod-level prometheus.io scrape annotations. The EKS clusters scrape with the
+  # Grafana Agent (→ Amazon Managed Prometheus) by pod annotation, with no
+  # prometheus-operator — so the ServiceMonitor above is consumed only on
+  # kube-prometheus-stack clusters (kx). These annotations make the operator's
+  # controller-runtime reconcile metrics reach AMP in prod. The operator
+  # NetworkPolicy already allows scrape from the monitoring namespace (where the
+  # Grafana Agent runs).
+  podScrapeAnnotations: true
 
 resources:
   requests: