JMIR Publications

ABSTRACT

In February 2025, Andrej Karpathy introduced vibe coding—building software by describing intent in natural language rather than writing precise code. This concept captured a broader paradigm shift: from prompt engineering toward context engineering, where the richness of context supplied to artificial intelligence (AI) determines output quality more than the precision of commands. We propose that the same principle applies to health. Vibe Health is a dual-sided paradigm in which individuals reach actionable health decisions through honest, iterative conversations with AI—without requiring medical knowledge or prompt expertise. The term deliberately extends Karpathy’s metaphor: just as vibe coding showed that programming skill matters less than the ability to articulate intent, Vibe Health posits that medical knowledge matters less than the ability to describe what is happening in one’s body. On the patient side, the core principle is that an honest prompt outperforms a perfect prompt: candid descriptions of symptoms, emotions, and lived context generate more useful AI responses than technically polished queries. On the physician side (Vibe Clinical), we reframe doctors not as novice prompt engineers but as the most experienced context engineers in any professional domain—their history-taking, physical examination, and clinical reasoning skills are precisely the context injection capabilities that enable high-quality AI interaction. We introduce the FTCAV model (Feel–Tell–Converse–Act–Verify) as an integrated behavioral framework that operationalizes Vibe Health for both patients and physicians. The model is grounded in interoception research—specifically the distinction between interoceptive accuracy (detecting bodily signals) and interoceptive awareness (reporting them)—and extends health behavior theory (the Capability–Opportunity–Motivation–Behavior model) to the conversational dynamics of AI-mediated health interactions. Each stage represents a discrete behavioral step: sensing a bodily signal or clinical cue (Feel), expressing it in natural language or structured clinical terms (Tell), refining understanding through multiturn AI dialogue (Converse), converting insight into executable action (Act), and confirming with appropriate authority (Verify). We examine the emerging medicolegal implications of AI-mediated health conversations, arguing that patients’ timestamped, contextualized AI dialogue logs carry evidentiary weight that physicians cannot safely ignore. We call for three specific actions: incorporation of Vibe Health principles into patient-facing AI platforms and health education programs, piloting of Vibe Clinical modules in medical school curricula, and development of professional guidelines for the documentation and clinical integration of patients’ AI conversation records.