Doximity bleeds on AI costs 📉, 65% of doctors use this AI tool 🩺, Abridge hits 100M visits 🎙️
Two-Thirds of US Doctors Are Now Using OpenEvidence — And It’s Funded by Pharma Ads
NBC News reported this week that approximately 650,000 US physicians — roughly 65% of all doctors in the country — actively use OpenEvidence, an AI-powered clinical decision support tool.
In April 2026 alone, the platform handled nearly 27 million clinical queries. On March 10th, it hit a milestone: one million clinical consultations in a single 24-hour period.
This is the largest clinical AI adoption event in history, and it happened without a single hospital IT committee voting on it.
The platform is free. It requires NPI verification, so only licensed clinicians get in. It cites peer-reviewed literature. And 60% of searches are about active clinical decisions — not board prep, not curiosity. Real patients, real choices.
The business model: pharma and medical device advertising. The same funding structure that built UpToDate’s early competitor landscape is now underwriting the AI layer that sits between a physician and their next order.
Here’s the part that should keep builders up at night. OpenEvidence raised $250M in January at a $12B valuation — roughly $700M total from Google Ventures, Nvidia, Sequoia, Mayo Clinic, and others. The adoption is real. The funding is real. And the question for every clinician-builder is: what do you build when two-thirds of your colleagues already have a free AI in their pocket?
😤 “It’s just a fancier UpToDate.” Have you used it? I think there is an AI UpToDate but it is 💰💰💰💰.
😤 “Pharma ad funding will bias the outputs.” This is the right concern and the wrong frame. UpToDate has pharma relationships too. The question isn’t whether ad funding exists — it’s whether the ad layer influences the clinical reasoning layer. That’s an auditable architecture question, not a hand-wave. If you’re a builder, this is an eval you can actually build and publish.
😤 “Doctors don’t need another AI tool.” 650,000 of them disagree. Adoption at this scale, without institutional mandate, means the tool is solving a real workflow pain point. The signal isn’t that doctors should use AI — it’s that they already are, and the rest of the ecosystem hasn’t caught up.
💡 80/20: The “let me look in UpToDate“ has shifted under your feet.
Abridge Crosses 100M Doctor Visits — From Scribe to Healthcare OS
Abridge, the clinical documentation AI company, has now processed over 100 million doctor visits across more than 250 health systems including Mayo Clinic, Johns Hopkins, and Kaiser Permanente.
A deep Latent.Space episode with co-founders Shiv Rao and Chai Asawa revealed the real play: Abridge is evolving from ambient scribe into a platform that handles prior authorization in minutes instead of hours.
The scribe was the wedge. The OS is the product.
😤 “Every AI company claims they’re becoming a platform.” True. But Abridge has the distribution (250 health systems), the data flywheel (100M visits), and the $550M war chest. The prior auth play is the test — if they can cut PA time from hours to minutes at scale, that’s a real workflow transformation, not a slide deck promise.
💡 80/20: If you’re building ambient documentation tools, the market just told you what comes next — the note is table stakes, prior auth and referral automation are the premium. Build downstream from the transcript, not around it.
Doximity Stock Plunges 25% — AI Costs Real Money
Doximity shares hit a record low after reporting that AI compute costs are eating into margins. Q4 revenue was up 5% but adjusted EBITDA margin guidance for FY2027 dropped to 49% — down from 55% — as AI spending ramps.
The market’s message: “AI-powered” is no longer a premium. It’s an expense line.
😤 “Doximity’s problem is Doximity, not AI.” Partly true — the healthcare advertising market is soft. But the margin compression from AI compute is a sector-wide signal. Every health tech company adding AI features is absorbing the same cost curve. The ones who can’t show ROI per AI-dollar will follow Doximity down.
⁉️ “Do you really know anyone who uses Doximity every day?” Exactly.
💡 80/20: If you’re pitching a clinical AI product, your CFO slide just got more important. The era of “we added AI” as a growth narrative is over. You need per-unit economics: what does each AI inference cost, what does it save, and when does the margin flip positive? Build that spreadsheet this week.
WaPo: Companies Masqueraded as Providers to Capture and Sell Health Records
The Washington Post reported that companies allegedly posed as healthcare providers to access and sell digital health records for profit. Hospitals are now urging stricter vetting of data access requests.
😤 “This is a compliance problem, not a builder problem.” Wrong. Every clinician-builder who touches health data should care about the credentialing chain. When bad actors exploit HIPAA’s Trust Framework to extract data, the regulatory response tightens access for everyone — including legitimate builders using patient-mediated FHIR or Cures Act protections.
Healthcare Costs Hit 15-Year High — Employer Health Benefits Expected to Exceed $18,500 Per Employee
Mercer’s 2026 survey projects a 6.5-6.7% increase in employer health benefit costs — the highest since 2010. Average cost per employee will push past $18,500. Christina Farr’s Second Opinion newsletter noted that 59% of employers are making cost-cutting changes, up from 44% two years ago.
💡 80/20: This is the demand signal for every builder selling cost-reduction tools. When CFOs are panicking about a 15-year-high cost curve, your AI-powered utilization management, care navigation, or prior auth automation tool has a warmer reception than it had six months ago. Lead with the Mercer number in your next pitch.
Out-of-Pocket Wearable Hackathon: Six Clinical AI Prototypes Built in 36 Hours
Nikhil Krishnan’s Out-of-Pocket newsletter recapped a hardware hackathon that produced six clinician-relevant prototypes in 36 hours: a trauma transfer coordination tool (TXA), a CPR copilot (Lazarus), a postpartum monitoring wearable (Tamago), a fall risk assessment device (SteadyStep), a monitoring hub (Eternal), and an NPS tracker for healthcare (Hospitality Helper).
💡 80/20: A CPR copilot and a trauma coordination tool built in 36 hours. The barrier to building clinical hardware prototypes is collapsing at the same rate as software. If you’ve been thinking “I need a hardware team for that” — you don’t, or at least you don’t need one for the first prototype.
What are you building this week? Reply and tell me — I read every one.
— Kevin