clinicians.dev

clinicians.dev

A community of clinical experts super-powered with skills in agentic engineering. Our roots are in coding and clinical medicine. We know how to create and supervise LLMs writing Python, SQL, R, JS, HTML. We love startups, regs, policy, UI/UX, data viz, project management, informatics. We want to create the medicine of the next 100 years.

✉ Email for Slack invite → Slack → Substack ↓ Next event
today · the wire

The Wire

What you'll be asked today — and what to say back. A couple that caught my eye this morning; the rest are on the wire.

Epic shipped real-time prior auth five months early. Does it authorize anything?

Sergei Polevikov, AI Health Uncut · Aug 27, 2026

It checks. It does not authorize. The feature went live at four health systems against UnitedHealthcare, Aetna and Network Health, five months ahead of the January 2027 CMS deadline — and what it actually does is tell you at the point of order that a prior auth will be required. It doesn't submit one and it doesn't resolve one. That gap between "real-time prior auth" and "real-time prior auth check" is worth being precise about before someone writes a business case around it. Knowing an auth is coming beats finding out three days later from a fax, but the friction itself hasn't moved — it's just been notified earlier, which mostly benefits whoever was paying to process the submission. Meanwhile the RCM vendors are buying the resolution layer: one has acquired the prior-auth company assembled from a previous AI vendor's wreckage, its second such deal in ten months. The EHR is shipping the notification. Somebody else is shipping the fix. Nobody's selling the part where the auth stops being necessary.

HatersWe've run this play before. Meaningful Use also promised transformation and delivered attestation. The vocabulary arrives years ahead of the function, everyone books the win, and in a few years someone writes a paper about why prior auth still takes eleven days.

96% on the simple risk score. 64% on the one that changes management.

Stephen Ranjan · Aug 28, 2026

A model extracting the Padua VTE score — a handful of fields — hits 96 percent F1. Move the same model to the 40-item Caprini score and F1 falls to 0.64, with low-risk sensitivity dropping to 0.22, meaning it misses roughly four out of five patients it should be flagging as low risk. Nothing about the model changed. What changed is the number of places the documentation can get messy, and real documentation is messier than a clean benchmark. This is the failure mode that matters — not dramatic hallucination, but a model reading a note about leg bandages and quietly inferring total paralysis. And when the output's wrong, the checking burden lands on whoever's using it, not the vendor who benchmarked it. Before any extraction tool goes into a workflow, it's worth finding out which end of that range you're actually living on.

BuilderIf you're shipping extraction, publish the degradation curve, not the peak score. Field count and note quality are the real independent variables, and the buyer who's been burned once will ask for exactly that chart. Being the first to volunteer it is a moat.

Stop asking your platform for an API. Ask it for a primitive.

Brendan Keeler, Health API Guy · Aug 27, 2026

A primitive is raw read and write access to data and capability. An abstraction is a predefined function that encodes the platform's assumptions about what you should want. Both get called "the API." Only one gives you optionality. A system of record is already embedded in the workflow; a point solution has to earn its way onto every screen. That asymmetry means the platform is permanently incentivized to ship an abstraction covering the eighty-percent case — and once it does, every point solution built on top looks the same to the buyer, because they're all calling the same constrained function. The AI wrinkle sharpens it: when the cost of building software collapses, the advantage goes to whoever has the least left to build, and that's the incumbent sitting on the data. Practical version for Monday: when a platform offers you an integration, ask whether you can read and write the underlying resource, or only call the function they wrote. If it's only the function, you're not integrated. You're enrolled.

80/20The one question that matters in the vendor meeting: can I write back? Read-only access lets you build a viewer. Write access lets you build a product. Everything else in the integration doc is decoration on that one bit.

Lilly's data says tirzepatide saves $319 a month — and that's exactly why Medicare's Bridge math barely works.

Diabetes, Obesity and Metabolism, via HTN Daily Brief · Aug 27, 2026

The numbers work on a narrow margin, and it matters that the study is Lilly-sponsored. Tirzepatide users showed $145 PPPM lower total healthcare costs than controls initially, widening to $319 PPPM by twelve months. Medicare's GLP-1 Bridge program costs $195 PPPM all-in. So the offset only clears somewhere in the twelve-to-eighteen-month window, and only if the patient stays on the drug — which is the whole business case in one clause. Real-world persistence isn't great, and a member who discontinues at month seven has generated the cost with none of the offset. CVS and Walgreens have already filled roughly 100,000 Bridge scripts between them in under two months, so this stops being theoretical fast. The sensitivity that actually moves the answer isn't the savings figure. It's the twelve-month adherence rate.

Out-of-network pricing just became a software category.

Health Tech Nerds, HTN Daily Brief · Aug 28, 2026

A startup called Daffodil Health is pitching flat-fee, AI-enabled claim repricing and No Surprises Act dispute handling, explicitly against incumbents who charge a percentage of savings. Their CEO's word for those incumbents is "fat and happy," and the pricing model is the entire wedge — a vendor paid a percentage of the dispute is paid more when the dispute is bigger, which is a strange incentive to build a cost-containment industry on. The tailwind is dispute volume: one analysis puts cumulative arbitration costs at $22.4 billion. Cost-containment vendors don't usually get disrupted on features. They get disrupted when the buyer starts suspecting the vendor benefits from the problem persisting.

BuilderThe pricing model is the product decision. Any time you can charge a flat fee where the incumbent charges a percentage, you're not competing on features — you're competing on the buyer's suspicion that their current vendor benefits from the problem persisting. That suspicion sells itself.
Read the full wire ↗
learn by doing

Interactives

Companion experiences for the newsletter — don't just read about it, try it.

Browse all interactives →
from the substack

Builder's Briefing

The newsletter — what shipped this week, what it means, and what to build next. A couple of recent issues.

Read more on Substack ↗
Summer 2026 · Free · Virtual

What the Vibe

Saturday, May 30, 2026 · 12p–2p ET

Vibe coding changed everything. The barrier to building just disappeared. So what's the vibe? Is clinicians.dev dead? Does knowing how to code still matter when anyone can build? Four clinician-builders. Twelve minutes each. Show us what you made — vibe-coded, hand-coded, or somewhere in between.

Tentative speakers
Joe Izzo
First, Do No Harm — Especially With Your MVP: A CMIO's Advice in the Clinician-Builder Era
Patrick Damaso
Trajectories, Not Targets: Catching Reward-Hacked Diagnostic Reasoning One ICD Branch at a Time
Sam Wang
Silent Denials, Loud Insights: Visibility into the Payors Who Won't Show Their Work
Matt Segar
From Bedside to Buyout: An EP Fellow Path to Acquisition
  • 4 talks (12 min each)
  • 1:1 breakouts after each
  • Group discussion
  • Open community talk: "What the Vibe?"
RSVP on Luma →
01 · Learn

AI & agentic engineering

The foundational reads, the tools clinicians-who-code are actually using, and the loop that ships.

Context engineering — the foundation

Effective Context Engineering for AI Agents

Anthropic's foundational guide.

Building Effective Agents

When to use workflows vs. agents, and how to structure both.

2026 Agentic Coding Trends Report

How teams are actually shipping with agents.

Learn CLAUDE.md / AGENTS.md

The spec files that make agentic coding repeatable.
🔁 The Ralph Wiggum Pattern — the loop that ships. Tight spec → agent implements → you verify → adjust spec → repeat. Just keep going.

Tools of the trade

ToolBest forNotes
Claude CodeComplex multi-file work, architectureTerminal-native, large context. The power tool. Good for non-code tasks too.
CursorDaily coding with autocomplete + inline editsAI IDE, huge user base.
WindsurfBudget entry, Cascade collaborationAI-native editor.
Copilot / Codex CLIRepo-native GitHub workflowsTight GitHub integration.

Skills marketplaces

Claude Code Marketplace

Curated agent skills collection.

Claude Skills (232+)

Cross-agent skills for Claude Code, Codex, Gemini CLI, Cursor.

Awesome Claude Plugins

Plugin adoption metrics.

claudemarketplaces.com

4,200+ skills, 770+ MCP servers.

Healthcare MCPs

AWS HealthLake MCP

Amazon's healthcare data MCP server.

Agent Care

EMR integration with FHIR for agentic AI.

Keragon

300+ native healthcare integrations, FHIR protocol support.

Superpowers Framework

Agentic skills framework for coding agents.

Agentic frameworks & orchestration

  • CrewAI— multi-agent orchestration, clinical workflow pipelines
  • LangGraph— graph-based agent workflows with state
  • AutoGen— Microsoft's multi-agent conversation framework
  • MS Agent Framework 1.0— production .NET + Python
  • DSPy— programmatic LLM pipelines, good for clinical NLP
02 · Communities

Communities, courses, substacks

Where clinician innovators and clinician builders are hanging out.

X Primary Care

Substack.

Automate Faculty

AI.

HealthTechNerds

Active Slack, all different backgrounds.

Tuva

Active Slack community.

CodeRx

Pharma, healthcare, technology.

Mimilabs

Medicare data (paid).

DiMe

Research, clinical, digital medicine.

Design for Healthcare

UX, UI, healthcare products.

HealthDevHub

FHIR.

HealthTech Hang

Networking + resources.

Physician Innovator

Entrepreneurship, medical innovation.

SoPE

Healthcare, entrepreneurship, VC (paid).
03 · Build stack

EHRs, FHIR, fake patients

The pipes you'll need when you're actually building.

EHRs & FHIR

🔑 SMART-on-FHIR / OAuth flow — fhir.epic.com/Documentation?docId=oauth2

Generate fake patients

MakeData

Synthetic healthcare datasets — FHIR, JSON, CSV. Privacy-safe, immediate.

Patient Creator GPT

ChatGPT custom GPT for fake cases.

Synthea

MITRE's synthetic patient generator — the classic.

FHIR Personas

Synthea patients curated for interesting characteristics.
04 · Datasets

Clinical datasets worth knowing

From CXRs to genomes to ICU waveforms. Filter by name, modality, or license.

NameRelevanceLicense
ReXGradient-160K160k multi-site CXR + reports — vision-language radiology sandbox.Harvard DUA, non-commercial
CheXpert PlusLarge paired CXR–report set; benchmarked in many papers.Stanford DUA, free research
Endoscapes 2023Open laparoscopic chole frames — segmentation / CVS detection.CC BY-NC-SA 4.0
Surg-3M3M surgical frames powering "SurgFM" foundation model.TBA, expect research-only
AFRICAI RepositoryImaging sets from African centres — fairness & domain shift.Mixed open licenses
OpenOximetryWaveforms + skin-tone data for pulse-ox bias work.PhysioNet credentialed
DeepLesion32k CT slices with bounding-box lesions; detection / tracking.NIH DUA, research-only
BioASQ Synergy 2024Biomedical Q-A pairs — LLM eval set.CC BY 2.5
CliniFactClinical-trial fact-checking corpus — fine-tune retrieval / RAG.MIT
Hallucination AnnotationsDoctor- & LLM-written discharge summaries with token-level labels.PhysioNet credentialed
Clinical-Trial Eligibility QAQA pairs linking MIMIC-IV to apixaban RCT criteria.PhysioNet credentialed
PIFIRWearable PPG/ECG for arrhythmia-free interval prediction.PhysioNet restricted
GREGoR R02Rare-disease genomic + phenotypic harmonised data.dbGaP controlled
Synthetic Rare-Disease EHRsBenchmark synthetic EHRs for low-prevalence conditions.CC BY
Korea4K4k Korean genomes — ancestry diversity for variant calling.EGA controlled
OpenNeuro20k+ public neuro-imaging sessions; BIDS-ready.CC0 / CC BY-SA
Bridge2AI-VoiceMultimodal speech (voice, vitals) for health AI.PhysioNet restricted
PMDB Pain MonitoringWearable IMU + self-report pain diary.CC BY 4.0
DREAMT Wearable SleepApple Watch PSG pairs for sleep-staging models.PhysioNet restricted
MC-MEDMulti-condition medical dialogue (GPT / human).PhysioNet credentialed
Wearable Stress DatasetSmartwatch vitals + stress labels — mental health ML.PhysioNet restricted
MIMIC-IV v3.1Flagship 380k-patient de-id EHR; ED + ICU tables.PhysioNet credentialed + CITI
MIETICItalian clinical-notes corpus with entity spans.PhysioNet credentialed
ODD (Opioid Behavior)Annotated notes for opioid-related behaviour NLP.PhysioNet credentialed
UK Biobank500k UK adult cohort — EHR, surveys, genetics.Controlled access
All of Us (NIH)1M-goal US cohort — EHR, surveys, genomics, wearables.Registered + Controlled tiers
TCGA~11k patients across 33 cancer types — multi-omics + clinical.Partially open
AmsterdamUMCdbFirst open European ICU DB — 23k admissions.DUA required
ADNILongitudinal Alzheimer's — serial MRI/PET, clinical, biomarkers.Free non-commercial
ABCD Study10k youths — neuroimaging, cognitive, mental health, genetic.NIMH controlled access
NHANES (CDC)US national survey — health, nutrition, lab data.Public domain
CheXpert (original)224k chest X-rays, 65k patients — labeled findings.Free non-commercial
EchoNet-Dynamic10k+ cardiac ultrasound videos with EF + ventricle volumes.Non-commercial
SyntheaRealistic synthetic patient records — full EHR.MIT
1000 GenomesWGS from ~2,500 diverse individuals — human variation reference.Open access
DementiaBank (Pitt)Speech recordings + transcripts from Alzheimer's patients + controls.Consortium access
VitalDB6,300+ surgeries with continuous high-freq vital sign waveforms.Open, registration + DUA
Medical Segmentation Decathlon10 open datasets for 3D medical image segmentation.CC BY-SA 4.0
PANDA10k+ prostate biopsy WSIs with Gleason grades.CC BY 4.0
05 · Conferences

AI & health conferences

Borrowed from Raihan Faroqui's healthcare AI resource guide. Filter to find your tribe.

06 · Jobs

Job boards that don't suck (as much)

What they are. Not endorsements — just the lay of the land.

07 · Startup

So you want to build a startup

But don't know anything about tech/stack, marketing/sales, finances, management.

1984 VC

Scrub Capital

Redesign Health

Flagship Pioneering

08 · Life

Life hacking / survival

Because clinician-builders need to not burn out.

09 · Experiments

LLM matrix tests (alpha)

Testing out LLMs with fake patient scenarios generated by GPTs.

LLM Rubric

GPT fake case #1

Sepsis markdown (Claude)

Healthcare AI resources

10 · Archive

Past events

Where we've been.

Fall 2025 Conference
TimeSpeakerTalk
1:00pTBDWelcome
1:10–1:22pJosh Mandel, MDConversational Interoperability for Prior Auth and Beyond — MCP, A2A, and the Unreasonable Effectiveness of Making Data Accessible
1:35–1:47pJung Hoon Son, MDData CPR: How LLMs Revive Buried Clinical Insights
2:00–2:12pJason Theobald, MDContain Multitudes: A Docker Intro
2:25–2:37pCalvin Johnston, MDSoftware Design and Family Medicine: Overlapping skill sets
2:50pTBDWrap-up
Summer 2025 Conference
TimeSpeakerTalk
1:00pKevin Maloy, MDWelcome
1:10–1:22pVishnu Ravi, MDDigital Health that Ships: An Open Framework You Can Use Today
1:35–1:47pAlex Dummet, MDTopic TBD
2:00–2:12pPawan Jindal, MDStartup Survival 101: A Clinician's Guide
2:25–2:37pOmar Usman, MDChange Data Capture (CDC): A Defiant Approach
2:50pKevin Maloy, MDWrap-up
Fall 2024 Conference (Nov 9, 2024)
TimeSpeakerTalk
1:00pKevin Maloy, MDWelcome
1:10–1:22pPaulius Mui, MDFrom 0 to 1: Becoming a Clinician Who Codes
1:30–1:42pBrian Fung, PharmDPath of the Clinician Engineer
1:50–2:02pJoe Izzo, MDThe Coding CMIO — it's okay to get your hands dirty
2:10–2:22pKarambir Khangoora, MDLearning to Code as a Clinician: Building Medagogy
2:30–2:42pJoey LeGrand, PharmDOne Brain Can Be Better Than Two
2:50pKevin Maloy, MDWrap-up
Podchat — Should Hospitals Mandate GenAI Training? (Aug 29, 2024)

Listen to a short podcast, then talk about whether it's right or wrong. Original from Bill Russell at This Week Health. → Meetup event