Diagnosing a child's sleep problems is too personal and high-stakes for AI to handle alone.

Every child's situation is unique — different temperaments, family stressors, medical histories, and environmental factors mean no two cases are structurally the same. The agent must exercise fresh, individualized judgment every time, which undermines automation value.

Success criteria are deeply subjective and hard to verify: did the intervention actually work, and was it safe and appropriate for this specific child? There is no crisp, observable signal of completion that a non-human can reliably evaluate.

The agent has no access to the child's medical records, behavioral history, school reports, family context, or real-time observations. All the data that matters is locked in the lived experience of the family and cannot be retrieved via API or file.

A misdiagnosis could mask a serious underlying condition — sleep apnea, anxiety disorder, trauma response, or neurological issue — delaying proper treatment and potentially causing lasting harm to a child's health and development.

This task demands clinical intuition, empathetic interviewing of a child and parents, ethical responsibility, and the ability to weigh nuanced behavioral cues — none of which current AI agents can reliably provide in a high-stakes, individualized medical-adjacent context.