I wish I learned to make chips instead of going to med school - Cerebras 🤑, OE as a standard?, Clara sells AI primary care for $25 💊
The Shadow Standard? Claiming OE Became the Standard Care Without Permission
A practicing clinician wrote this week: OpenEvidence has quietly become the de facto “standard of care” for clinical decision support — without formal adoption, without oversight, without disclosure.
Reminds me of Butterfly. One day nobody had a handheld ultrasound. Then suddenly they could have one in their coat pocket, and committees were scrambling to write policies for a device that was already being used on patients. No RFP. No formulary review. Just adoption that outran governance.
OpenEvidence is the same pattern, except it’s shaping clinical reasoning instead of clinical imaging. Physicians use it to prep talks, chase differentials, look things up between patients. It’s not “standard of care” in the legal sense - people are using it but I don’t think on every case. But it’s everywhere and invisible.
BTW - if you want to give into legalese stuff, highly recommend Gita Pensa’s The L Word podcast … better than reading substacks about “standard of care.”
😤 “65% doesn’t mean standard of care.” Correct. It means 65% used it at least once in a month. That’s adoption, not entrenchment. I wonder how uptodate’s traffic has fallen off.
⁉️ Enough covering OpenEvidence: You are right, we need a moratorium. It’s kinda like talking about over-coverage of pulmonary embolism on an emergency medicine podcast.
Clara: $25/Month AI Primary Care from the Circle Medical Alumni
Clara launched out of YC with a $12M raise led by A.Capital and a pricing model that prices out every Forward, One Medical, and Carbon Health membership tier.
Three tiers: $25/month (chat, prescriptions, labs), $50/month (adds wearable integration and 74 biomarker tests), $150/month (108 biomarker tests, hormone panels). Every prescription and lab order is physician-reviewed. HSA/FSA eligible. The founding team — Favvas, Swift, Ahmed — built and grew Circle Medical to $100M in revenue before this.
This is the cleanest 2026 instance of the Forward-redux pattern at consumer scale.
😤 “$25/month AI primary care will be a race to the bottom.” Maybe. But the founders already scaled one primary care company to $100M. The question isn’t whether $25 is sustainable — it’s whether the unit economics work at the Hims & Hers price band. If they do, every membership-based primary care startup just lost their pricing floor.
😤 “Physician-reviewed doesn’t mean physician-quality.” Fair concern. But the architecture matters here: FHIR-aggregated records at 150K+ user scale, physician sign-off on every order. The quality question is empirical, not structural. Watch the outcomes data.
💡 80/20: If you’re building consumer health tools, Clara just set the price the market will compare you against. $25/month for AI-powered primary care with real physician oversight is the new benchmark. Your product better deliver something Clara can’t — specialty depth, chronic disease management, or institutional integration.
Photon + Doximity: E-Prescribing Goes PLG
Photon Health announced a partnership with Doximity to bring e-prescribing into Doximity’s clinical AI suite for independent practices.
Over 1,000 prescribers participated in the beta. Photon raised $16M in Series A for modern prescription infrastructure — transparent pricing, patient pharmacy choice, API-first architecture.
The play: a prescription API that sits underneath whatever clinical workflow tool a doctor uses.
💡 80/20: If you’re building clinical tools that touch prescriptions, Photon’s API is the infrastructure play to watch. The e-prescribing market is finally getting the API-first treatment that payments got five years ago. Build on the API, not the form.
ACA Enrollment Crater: 21% Non-Effectuation Rate Signals Marketplace Contraction
Preliminary CMS figures show non-effectuation rates hitting ~21% — up from 12% last year — meaning the marketplace denominator collapsed from ~21.4M to ~19M effectuated lives.
The driver: ARPA enhanced-subsidy expiration at end of 2025. That’s an 11% absolute contraction in the marketplace-insured population.
💡 80/20: If more than 25% of your customer base is marketplace-concentrated, run a sensitivity analysis this weekend. A TAM that shrank 11% overnight changes your growth model.
Cerebras IPO Closes at $60B — Inference Infrastructure Gets Its Moment
Cerebras went public at $60B ($280/share), validating wafer-scale AI hardware. CFO Bob Komin disclosed they serve trillion-parameter OpenAI models (5.4, 5.5).
Six months after NVIDIA acquired Groq for $20B, the inference infrastructure market just got priced.
💡 80/20: For health tech builders, this matters because inference cost determines what clinical AI applications are economically viable. When inference gets cheap enough, the clinical reasoning agents that are currently too expensive to run at scale become table stakes.
CMS Final Rule Lowers ACA Exchange Costs, Expands State Control
CMS published a sweeping final rule for ACA exchange plan year 2027: lower user fees (1.9% for FFE, down from 2.5%), tighter eligibility verification, and expanded state authority over plan oversight and certification reviews.
💡 80/20: States getting more control over exchange plan certification means the regulatory landscape fragments further. If you’re building for payers or exchanges, expect state-by-state variation to increase — and plan your compliance accordingly.
🛠️ From the Workbench
A Clinician’s Field Guide to Claude Cowork’s Three Modes
A practicing surgeon published a practical map of agentic work for clinician-operators: when to dispatch a one-off task, when to schedule recurring automation, and when to leave a live artifact running.
This is the kind of guide that only works when written by someone who actually has clinical workflow constraints — 12-hour shifts, pager interruptions, no time for tinkering during patient care.
⚠️ Verify: Claude Cowork is an Anthropic product currently in research preview. Any clinical workflow automation involving patient data should go through your institution’s IT security and compliance review before production use.
💡 80/20: If you’ve been curious about agentic AI tools but haven’t tried them in a clinical context, this guide gives you the three-mode mental model. Start with dispatch mode on a weekend project using synthetic data. Graduate to scheduled tasks when you find a pattern worth repeating.
What are you building this week? Reply and tell me — I read every one.
— Kevin