I include the agent on images because, hopefully, it ensures the system will become managed immediately and it doesn’t rely on staff pushing the agent to the PC subsequently or figuring on some mechanism to automatically install the agent post image.
I install the agent at the end of the image creation just prior to sysprep, unplugging the network cable first, so the agent can’t contact the NS server at all, assigning a GUID etc, I don’t know if that is the best way to do this but it seems to work ok.