I build software for three audiences who share one thing: a generic AI response will hurt them. Couples are planning their wedding, in the most emotionally loaded purchase of their lives. Families of missing people are in active grief and vulnerable to anything that sounds like false hope. And anyone capturing their own worst moments in a private journal, where a chipper AI cheerleader does real damage. The interesting design work in each of these production systems isn't what the AI is allowed to say. It's the words that are forbidden, the pauses that are forced, and the protected modes the system locks itself into when it notices the user isn't okay.

This talk walks through the actual code and prompt patterns from three live products: a Claude-powered email agent for wedding venues with an escalation list that routes high-stakes emails away from the AI entirely; a missing-persons intelligence platform with epistemic disclaimers banning words like "confirmed" and "matched" from every prompt; and a personal AI companion that auto-locks users into a body-image protection mode when their own captures suggest distress.

Attendees leave with a pattern language for designing AI for high-stakes audiences: when to forbid words at the prompt level, when to route around the model entirely, when to silently upgrade the model based on emotional signal, and when the right answer is silence. With production code, real metrics, and the failure modes that taught me to add each pattern in the first place.