We should (optionally) stall the sending of a message after a Letta agent has generated it based on how much time has elapsed and vs how long an average human would have taken to type the message. This helps prevent chats from flowing too quickly in a group setting (faster than you can realistically read).