JMIR Publications

ABSTRACT

Time constraints and fragmented pre-consultation information are persistent challenges in general practice. This viewpoint describes the development, deployment, and discontinuation of Sokrates AI, a physician-developed, open-source chatbot designed to conduct Socratic pre-consultation history-taking in Norwegian primary care. Built almost entirely through AI-assisted code-generation through natural language prompting (vibe-coding) using OpenAI Codex/GPT-4 and Cursor.ai, the tool was deployed on a GDPR-compliant server and used informally with approximately 20 patients over one month in autumn 2025. Clinical utility was most apparent for asynchronous e-consultations and for patients disclosing psychosocial concerns; the tool was ultimately discontinued due to implementation burden and the availability of simpler alternatives for routine use. The central argument of this viewpoint is that AI-assisted development has meaningfully lowered the threshold for clinician-led digital innovation, enabling a practicing physician with no formal software engineering training to build and deploy a functional clinical application. This shift has implications for how the medical profession engages with the AI systems entering clinical practice, and for how medical education should respond. Three further observations from the project—concerning clinical adoption barriers, the unsuitability of LLMs for triage, and the risk of digital inequity—are noted as hypotheses warranting dedicated investigation rather than conclusions this single experience can support.