Is Vibe Coding a 'Fire of Prometheus' for Non-Developer Clinicians, Especially Clinical Radiologists?

Abstract

Vibe coding is a workflow in which users describe the desired software behavior in natural language, following which large language models (LLMs) generate the code. It is increasingly being used by non-developer clinicians without any formal programming training to create tools that support the clinical workflow. This development is both timely and consequential for radiology, one of the specialties both most exposed to artificial intelligence (AI)-driven automation and best positioned to adopt it owing to its longstanding engagement with imaging AI, digital infrastructure, and information system workflows. Therefore, radiologists may become not merely passive AI users, but also clinical workflow orchestrators. This opportunity is illustrated by a structured prostate magnetic resonance imaging reporting tool, vibe-coded by a non-developer radiologist in approximately four hours, with an institution-based voluntary, radiologist-led coding community. This investigation highlighted four cautions: erosion of domain expertise, risks to security and patient data, absence of peer feedback, and uncoordinated proliferation of tools. A minimum safeguard set for clinician-built tools was proposed. As previously framed, LLMs represent a "fire of Prometheus" for non-native English-speaking researchers. Similarly, vibe coding may warm and burn radiologists equally, with its impact determined less by technology than by how responsibly the specialty adopts it.