Every instance in the fleet is a record on the control instance, and every change to it is an action — a Hosting/InstanceAction you create with the instance's name, what you want (Roll, Restart, Suspend, Reactivate, Audit, Reconcile, …) and a confirmation. The operator running inside the cluster does the rest and writes each phase back onto the same node, so you watch one page instead of a terminal.

What changed

Nothing in the cluster. What changed is the rule, and the documentation that states it: an az aks command invoke … / kubectl … line is no longer the procedure. Every page that used to open with one now names the action that does the same thing, and keeps the old command only as break-glass — for the case where the control plane itself cannot act — or as the measurement behind a war story.

What is honest about it

Three questions still have no portal answer today: which image and how many restarts each replica has, what the process logged at a given minute, and whether a particular replica can load a particular node type. The new page Operating from the portal, not the cluster says so, names the read-only break-glass shape for each, and is the page the work closing those gaps points back at.