Your Doctor Just Built an App — And They Didn’t Write a Single Line of Code – Agile Mindset & Execution

A cardiologist in Brussels won a hackathon. An emergency physician in San Francisco co-founded a clinical calculator used by millions. Neither of them started as a software engineer. Both built their tools using AI — and neither had to write a line of code.

Here’s something that might catch you off guard: the people redesigning how medicine works aren’t coming from Silicon Valley. They’re coming from hospitals, clinics, and on-call rooms. They’re physicians — often exhausted ones — who got tired of waiting for someone else to fix broken workflows and just… started building.

And the thing that made this possible? Large Language Models. AI that speaks plain English back to you. Tools that turn a doctor’s idea into a working application without a computer science degree in sight.

Wait, Doctors Are Actually Building Software Now?

Yes. And not just simple forms or PDFs. Real, functional clinical applications.

Dr. Graham Walker, an emergency medicine physician and co-founder of MDCalc — one of the most widely used clinical decision tools in the world — has been using Claude Code to extend that vision. Dr. Michal Nedoszytko, a cardiologist at Cliniques de l’Europe in Brussels, won a Claude Code hackathon building clinical solutions without an extensive coding background.

Anthropic — the AI safety company behind Claude — recently spotlighted this trend with a dedicated webinar series on how healthcare professionals are using Claude Code to build tools for patient care. Their core insight: the same capabilities that make AI effective for software development map surprisingly well to clinical and scientific workflows.

Think about it. Medicine already runs on structured logic. Treatment pathways. Diagnostic criteria. Risk stratification. These are decision trees. AI is very good at decision trees — especially when a clinician can describe them in natural language.

It’s Not Just Claude. Here’s the Full Landscape.

Multiple AI models are finding their footing in healthcare — each with different strengths:

Claude (Anthropic) — Particularly praised for nuanced reasoning and handling long, complex clinical documents. A 2025 study published in PMC found that Claude 3.5 Sonnet outperformed both junior and senior physicians in assisting with diagnosis and management of autoimmune diseases. Its ability to interpret complex clinical issues was described as particularly outstanding.

GPT-4 / GPT-4o (OpenAI) — The most widely adopted in clinical settings, used for everything from drafting patient letters to supporting diagnostic reasoning. Doximity’s GPT-powered tools help physicians compose insurance letters and generate patient education content. A 2024 radiology study compared GPT-4o, Claude 3 Opus, and Gemini 1.5 Pro on diagnostic cases — all performed impressively.

Gemini (Google) — Strong at multimodal tasks, making it relevant where imaging reports and clinical notes intersect. Google’s infrastructure gives it a natural pathway into hospital systems already on Google Cloud.

LLaMA (Meta) / Open-source models — For health systems wary of sending patient data to third-party servers, open-source models running on local infrastructure offer a privacy-first path. ClinOps AI uses Groq-powered Llama 3 for sub-second clinical inference with full audit logging.

The Shift That Nobody Told You About

Here’s the real story underneath all of this: AI in healthcare isn’t just being deployed at the enterprise level anymore. It’s being democratised — one clinician-builder at a time.

The American Medical Association found that physician AI adoption jumped from 38% in 2023 to 66% in 2024. That’s not marginal growth. That’s a tipping point. Over half of those physicians cited administrative burden reduction as the primary driver — they wanted to spend less time in the notes and more time with patients.

But a growing subset isn’t just using AI tools. They’re making them. Physician Eve Cunningham MD called 2026 “the year of the clinician-builder” — a new breed of healthcare professional who prototypes workflows, builds voice agents for clinical triage, and deploys risk scoring models without a dedicated engineering team.

What Are They Actually Building?

Clinical decision support tools — calculators and scoring systems that help clinicians make faster, evidence-based choices at the bedside
Ambient documentation systems — AI that listens to patient consultations and auto-generates structured notes, cutting documentation time dramatically
Patient education content generators — tools that translate complex diagnoses into plain-language explanations personalised for each patient
Risk stratification dashboards — real-time alerts for high-risk patients based on EHR data, with full audit logging
Preoperative education systems — GPT-4-Turbo, Claude 3 Opus, and Gemini 1.5 Pro have all been evaluated for generating informed consent content that both clinicians and patients find useful

But Is It Safe?

This is the question everyone should be asking — and the good news is, the clinician-builders are asking it too.

Safety, auditability, and compliance are front and centre in these conversations. Anthropic’s healthcare sessions explicitly cover how physicians are approaching output traceability — making sure AI-assisted decisions can be reviewed, questioned, and explained. Researchers at the University of Colorado’s LARK Lab frame it clearly: LLMs in medicine are there to support clinicians, not replace their judgment.

A study published in Nature (2026) found that among AI-trained physicians, access to an LLM substantially improved diagnostic reasoning without slowing case review — a signal that effective AI could help address diagnostic gaps, particularly in resource-limited settings.

The key safeguard? The clinician stays in the loop. AI surfaces the evidence; the human makes the call.

What This Means for the Future of Healthcare

The traditional model of healthcare IT — where a physician describes a problem, hands it to an IT team, waits 18 months, and receives something that doesn’t quite match what they asked for — is being disrupted from the inside.

When a cardiologist can describe a workflow problem in plain language and have a working prototype by end of week, the dynamic changes. When an emergency physician can build a clinical calculator used by millions without hiring a development team, the barrier between idea and impact collapses.

AI isn’t replacing doctors. It’s turning the best ones into builders. And that might change everything about how the next generation of healthcare tools gets made — by the people closest to the problem, with the deepest understanding of what actually needs to be solved.