Your colleague uses 7 AI tools daily 🩺, hospital prices outrun Medicare 💰, patients get their own scribe 📱

Seven AI Tools, One Clinical Day, Zero Surprises

Doug Fullington, MD — a practicing internist — published a walkthrough this week of every place AI shows up in a single clinical day in May 2026.

The list: an agentic workflow platform that assembles pre-visit summaries overnight, cutting complex chart prep from nine minutes to three. An ambient scribe that produces signable notes roughly four out of five times, eliminating post-visit charting. AI-assisted coding that surfaces diagnostic codes and service levels at sign-off, reducing downcoding appeals. OpenEvidence at the point of care during visits — with the laptop turned toward the patient to review evidence together. Microsoft Copilot 365 drafting portal replies and lab-result letters, returning approximately 45 minutes per clinic day. A custom Claude project delivering structured clinical evidence briefings three mornings a week.

Seven tools. One internist. One day. And the number didn’t surprise a single person in the replies.

That last part is the story. Two years ago, a clinician using one AI tool was a conference keynote. Today, seven tools stacked into a single workflow is just ... Tuesday. The adoption didn’t follow the hockey stick. It followed the fog — you didn’t notice it arriving until you couldn’t see without it.

The cumulative time savings matter. But the architectural pattern matters more. No single tool here is transformative. The stack is transformative. Chart prep feeds the ambient scribe. The scribe feeds the coding engine. The coding engine’s output feeds the billing cycle. The evidence tool feeds the clinical decision. The portal tool feeds the patient relationship. Each tool makes the next one more useful — and harder to remove.

😤 “Seven tools is fragmentation, not progress.” It’s both. The stack is fragmented because no single vendor owns the full clinical workflow yet — and that’s actually the builder opportunity. The clinician who maps the integration points between these seven tools understands something the vendors building each individual tool don’t: how they interact in the real workflow. That’s architectural knowledge with commercial value.

😤 “Most doctors aren’t using seven AI tools.” Most doctors aren’t counting seven AI tools. The ambient scribe is “just documentation.” The coding engine is “just billing.” OpenEvidence is “just a search.” When you stop calling them AI and start calling them workflow, the number goes up, not down.

😤 “This is one doctor’s experience. It’s not generalizable.” We covered the OpenEvidence numbers last week — 650,000 physicians, 27 million queries in April. This isn’t one doctor’s experiment. It’s one doctor mapping what most clinicians are already doing without a name for it.

💡 80/20: If you’re building clinical AI tools, the competitive landscape isn’t other AI tools. It’s the stack. Build for the stack, not for the demo.

---

KFF: Hospital Prices Grew 47% Faster for Private Insurance Than Medicare

KFF published data showing private insurance prices for hospital care rose 30% from April 2019 to April 2026, compared to 21% for Medicare — a 47% faster growth rate. Private insurance payments now average 199% of Medicare rates overall, and 264% of Medicare rates for outpatient services.

The gap between what hospitals charge private insurers and what Medicare pays is the single largest structural tension in US healthcare economics. It’s also the reason every value-based care model, every direct primary care play, and every employer health plan is trying to route around the hospital.

😤 “This has been true for decades. It’s not news.” The rate of divergence is the news. Private prices grew at a similar pace as Medicare from 2019-2020, then accelerated every year from 2020-2025. The gap is widening, not stable. If you’re modeling unit economics for a clinical tool, the price assumption you use — Medicare or commercial — changes your TAM by 2-2.6x.

💡 80/20: If your product saves hospitals money on the commercial side, your ROI story just got stronger — the per-unit savings are 2x larger than the same intervention on the Medicare side. If you’re building for employers or direct contracting, the 264% outpatient gap is the number that makes CFOs listen.

---

Kin Health Raises $9M to Build the Patient-Side Scribe

Kin Health raised $9M led by Maveron to build what it calls the first consumer health platform designed around the physician-patient conversation. The app records medical visits and turns them into summaries patients can act on and share. Free. Always.

Founded by Arpan and Amit Parikh — practicing physicians — and Kyle Alwyn, who built HeyDoctor (acquired by GoodRx). More than 30 physicians invested in the round.

If 75-90% of health systems now have ambient scribes for providers, Kin is asking the obvious question nobody answered: who’s scribing for the patient?

😤 “Patients recording visits creates liability concerns.” Yup, check your policies about patients recording audio during a visit.

😤 “Free consumer health apps don’t survive.” The GoodRx playbook survived. The investor roster — including the GoodRx co-founders — suggests Kin has a monetization thesis beyond the free tier. The distribution play is what matters: if patients start bringing Kin summaries to their next appointment, the physician’s workflow changes whether or not the physician chose it.

💡 80/20: If you’re building patient-facing tools, Kin just defined the interaction model — record the visit, summarize it, make it actionable.

---

AMA Survey: 40 Prior Auths Per Week, 94% Say It Causes Burnout

The AMA’s 2026 Prior Authorization Physician Survey dropped fresh numbers from 1,000 physicians: 40 prior authorizations per week. 13 hours of physician and staff time weekly. 32% of requests denied. 94% say PA causes burnout. 26% report PA caused an adverse patient event.

And the number that should concern every prior auth AI builder: 60% of physicians are concerned AI will increase denial rates.

😤 “AI prior auth is supposed to fix this.” It depends on which side of the transaction the AI sits on. If you’re building AI that helps payers auto-deny faster, you’re building into the 60% physician resistance number. If you’re building AI that helps physicians submit better-documented requests that survive review, you’re building with the workflow. The product framing matters as much as the technology.

💡 80/20: The 40-per-week, 13-hours-per-week numbers are your pitch deck slide one.

---

Knit Health Launches from Berkeley with $11.6M and a Different AI Architecture

Knit Health emerged from stealth with $11.6M (Uncork Capital, Frist Cressey Ventures) and a “Large Clinical Behavior Model” trained on Truveta EMR data from 130 million patients across 30 US health systems. Instead of training on medical literature, Knit trains on what clinicians actually do — using behavioral cloning and causal inference to learn decision patterns.

💡 80/20: Most clinical AI is built on what medicine says. Knit is building on what clinicians do. The gap between those two things is the entire field of implementation science. Watch whether the behavior model outperforms literature-trained models on triage and patient flow — if it does, the architecture is the differentiator.

---

Oscar Health Q1: EPS $2.07 vs $1.11 Expected, Stock Pops 10.6%

Oscar Health beat earnings expectations with $2.07 EPS versus $1.11 consensus, $4.65B revenue (52.7% YoY growth), and a 70.5% medical loss ratio. The stock jumped 10.6%.

💡 80/20: Oscar is the proof point that tech-forward insurance models can achieve operating leverage at scale. For builders selling into health plans, Oscar’s results make the “technology reduces MLR” argument more credible in your next payer pitch.

---

WSJ: Nurse Practitioner Is Now the Hottest Job in Healthcare

The Wall Street Journal reported that nurse practitioners are the hottest job in healthcare. NPs have ranked #1 in US News’s Best Jobs for three consecutive years, with 40%+ projected employment growth through 2033 and median income of $132,050.

💡 80/20: If you’re building clinical workflow tools, the NP is increasingly your primary user — not the physician. NP-specific workflows differ from physician workflows in scope, supervision requirements, and documentation patterns. Build for the user who’s actually growing.

---

🎙️ From the Pods

“500-700 Basis Points Coming Out of Healthcare GDP” — Eric Larsen on The Heart of Healthcare

Eric Larsen — president of TowerBrook Advisors, venture partner at Thrive Capital, and writer of “Gen AI Juggernaut” — returned to The Heart of Healthcare with a framework every builder should internalize.

The concept to remember: “functional verifiability.” Larsen borrows from Karpathy: in software 1.0, if you could specify the function, you could automate it. In software 2.0, if you can verify the function, you can automate it. Revenue cycle management, coding, prior auth, billing — these are functionally verifiable. That’s why they automated first.

Larsen’s prediction: the next chapter is deflationary. Once payers match providers on AI capabilities, the coding-optimization arms race cancels out. Then the real deflation begins — not from better billing, but from labor substitution in the 23.8 million jobs that healthcare currently employs.

💡 80/20: The “functional verifiability” test is your product-category compass. If the output of your AI tool can be verified against a ground truth (correct code, correct claim, correct diagnosis), the automation case is strong. If it can’t be easily verified (treatment planning, care coordination, clinical judgment), the human-in-the-loop isn’t going away — and your product design should reflect that.

---

What are you building this week? Reply and tell me — I read every one.

— Kevin