I was thinking it would be interesting to get the model to generate timestamps with the messages. Then you could actually queue the messages until that time. It would be like a real conversation.

Of course if you send a message before the AI does it re-runs and produces a new future message with a new timestamp.