(Translated main content in English, with all Korean practice problems preserved below)
After 6 months of CloudFormation use, one weakness emerges: Run Command is one-time execution. Restart nginx fleet-wide today; 30 minutes later new ASG instance boots untouched. CloudWatch Agent deployed everywhere; next week someone deletes it from one instance, that one silently drifts. Fleet is living organism; one command doesn't maintain state.
State Manager solves with "desired state" abstraction. Not "run this command once" but "this state must always be true," let SSM periodically reconcile actual to declared. Deep connection to Kubernetes/Puppet/Chef desired-state models through this lens.