How would you design the mobile client for token-by-token AI response streaming so the UI stays responsive at xAI?
