There are different levels of effemeriality. The simplest example I use daily would be an autoscaling group in AWS. Especially if you use Spot Instances to save money, thi gs may scale in and out whenever.

So if a development team creates a new autoscaling group and I need to get into an instance to test something, unless I add stuff to their IaC, I'm stuck with their configuration. I need to assume that every time I ssh into one of those instances, it's a brand new instance. But it'd be a big challenge for me to go to their repo and make a PR to alias a command whenever an instance in that resource is created

Stuff can be even more temporary if it's something like an ECS task which creates a container with a read only filesystem only when a task is needed to be done. But I don't want to get too deep in the weeds (or deeper than I already have)

terraform workspace will at least stick around for a while so you might be in and out of the same system multiple times.