I think prefect 2.0 docs are off, there is line the about work-queues and agents section that states this:
“You must start an agent within the environment in which it will execute flow runs.”
This statement is a little off and should come with note stating the agent doesn't necessarily have to be running in the same execution environment as the flow. For instance you can have an agent running in a VM connected to a k8 cluster using kubectl and run k8 jobs that way.