AI companions for kids are shipping fast, with always-on, always-agreeable personalities. What developmental risks worry you most - attachment, social skill atrophy, or something else - and what guardrails would you actually enforce?