I'm imagining an accumulator variable that helps dictate AI diplomatic behaviour for purposes of this post, but I understand that's nothing but an analogy.
It would be very, very cool if messages themselves had diplomatic consequences, both direct and indirectly related. In our current example, let's say you send the AI a message to vacate the premises with his spy ship immediately. This involves a small relationship hit. The AI decides to linger for whatever reason instead of leaving post-haste. You destroy his spy. This is another relationship hit, but the two hits together are not as extreme as if you had fired immediately without diplomatic communication.