Only a single Agent instance can run on a machine, even on a server that allows multiple RDP connections. When a test is running, the browser process is spawned under the same session where the user started the Agent (e.g. session #1). When another user in another session (e.g. session #2) starts a test, it will run in the first session, and the user won't see it.