This change https://github.com/jenkinsci/remoting/pull/193/commits/69f9ebe72c608e14745fb3fc5f8d9e4c65758d89 is modifying the behaviour of the agent connection cycle.

Retrying the connection even when the agent is not yet created in the master was something a cloud implementation could rely on (one of our proprietary implementations does).

IMHO this was a breaking change. There should be at least one way to keep the previous behaviour (with a new parameter, or maybe reusing the existent "noReconnect").