Applications run as Deployments, not as bare Pods. A Deployment holds a Pod template and a replica count. It creates a ReplicaSet, and the ReplicaSet keeps exactly that many Pods from the template running: if a Pod crashes, is deleted or loses its node, a new one replaces it on a healthy node.
kubectl create deployment NAME --image=IMAGE --replicas=N makes one quickly; real teams keep the manifest in a file. The Pods get generated names, the Deployment's name plus the template's hash plus a short suffix, which is why you select them by label rather than by name.
kubectl scale deployment NAME --replicas=N changes the count. Always scale the Deployment, not its ReplicaSet: the Deployment owns the ReplicaSet and would put the old number back. The template's labels must match the Deployment's selector, or the API rejects the object.
A Deployment, the ReplicaSet it owns and the Pods that belong to it.