agents can only ever be grey or white boxes

A hypothesis I have: there's no way to define an "agent" that makes no reference to agentlike behaviour. ergo, we will always know things about an agent.