a16z bets $30M on the phone call 📞, K Health and endocrine 👀, Scribes code what you didn't say 🎙️
a16z just put $30M behind the phone call — and that tells you where healthcare AI is actually shipping
Andreessen Horowitz led a $30M Series A into Prosper AI (Base10 in; Emergence, Y Combinator, Company Ventures following on), first reported by Axios. The product is agentic voice AI that schedules the appointment, verifies the benefits, and clears the patient’s financial responsibility — across 150,000+ providers.
Here’s the part that matters more than the logo on the term sheet: revenue grew 5x in six months, the platform now touches $1.3B in patient care, and the company says it wins 80% of competitive bake-offs.
While everyone argued about which model answers a clinical question best, the durable money quietly moved to the back office.
The money in healthcare AI is migrating from clinical to administrative, and the org chart is moving with it. The fastest “yes” in healthcare lives in the administrative budget — CFO-owned P&L, hard-dollar ROI, no governance committee with four chairs that can each veto you.
A clinical decision tool has to clear the CMIO, the CISO, quality, and legal. A tool that cuts admin cost 40% and lifts scheduled visits 12% clears the person whose number it moves. That asymmetry is the whole game right now.
😤 “So it’s a fancy call center. Groundbreaking.” The call center is a $1T/year administrative drag and the single largest source of patient leakage before care even happens. “Boring” and “where the money is” are usually the same sentence in this business.
😤 “Voice agents calling payers will break the second a real edge case shows up.” Probably — at the edges. But eligibility and scheduling are the most scripted, most repetitive surface in the building, which is exactly why this is the part agents scale in production first. You don’t start autonomy at the 2 AM potassium. You start it on hold music.
😤 “a16z money doesn’t mean it works.” Fair. Watch the named go-live dates and the phone-verifiable metric, not the valuation. That’s the difference between this and a benchmark score.
❓ If the back office is where agents are actually going live, what’s the clinician-built tool that rides the same rail — the thing that uses the eligibility/scheduling data nobody else is sitting on to do something clinical the CFO will still happily fund? I think there’s a wedge there and I can’t quite name it yet.
Epic can host, but it still can’t sell small — and that’s your opening
Brendan Keeler makes the case that Epic’s integrated-monolith architecture is exactly why it dominates acute care (43.7% share) and exactly why it keeps harpooning at the small-practice “white whale” and missing.
His verdict on Garden Plot, Epic’s downmarket product: ~29 customers, almost all single-site across two states, no growth in nine-plus months — versus 162 Epic Hosted tenants. The failure is go-to-market, not technology.
The independent-practice segment that the biggest EHR on earth has chased for a decade and still can’t crack is the segment a clinician-builder actually understands.
😤 “Athenahealth and eClinicalWorks already own that market.” They own the EHR. They don’t own the dozen specific workflows a solo practice hates — and that’s a thinner, friendlier wedge than competing with Epic head-on.
💡 80/20: “Too small for Epic to bother with” is a market description, not an insult. Build the narrow thing for the practice Epic’s salesforce will never call back.
The RPM point solution is getting eaten by the delivery layer
ChartSpan acquired Validic, folding a normalized device API spanning 700+ consumer and clinical-grade devices (~20 million connected lives) into a full-service chronic-care-management team.
The pitch is a single layer that turns patient-generated data into billed clinical action instead of a dashboard nobody acts on.
The durability lesson for builders: the standalone point solution you pilot this quarter can be a feature inside someone’s care-delivery stack before you scale it.
😤 “Device-data plumbing is a commodity.” The plumbing is. The part that’s hard — and acquirable — is the clinical team that turns a glucose trend into a CCM/APCM claim. Own the action, not the API.
💡 80/20: If your tool’s only moat is “we connect to the devices,” assume that’s a line item in someone’s next acquisition. Move up the stack toward the decision and the billing event.
The AI front door is walking into the specialty hallway
K Health is extending its Hartford HealthCare partnership beyond primary care into specialty, starting with endocrinology — an early case of AI triage moving past the front desk and into the subspecialist’s queue.
Endocrinology first is not an accident: it’s protocol-heavy, data-rich, and chronically short-staffed — the friendliest possible beachhead for AI-assisted intake. Watch whether the model that triages a thyroid panel can survive the messier specialties next.
Ultra-short:
The “doctor shortage” might be a distribution problem, not a supply one. A widely-shared analysis points out Boston has the most physicians per capita in the country and some of the longest appointment waits — which is a scheduling-and-access problem builders can touch, not a med-school-class-size problem they can’t.
MCP just got enterprise auth. The Model Context Protocol shipped Enterprise-Managed Authorization — zero-touch OAuth so IT can govern which agents reach which systems.
🎙️ From the Pods
🎙️ Health Tech Nerds Radio — “The Grand Roundup”
The unsettling builder pearl: ambient scribes quietly make billing and clinical determinations out of ambiguous chat. “I get dry mouth sometimes” becomes coded xerostomia; a passing end-of-life remark gets logged as a documented advance-care-planning conversation the clinician says never happened.
🔇 Speaker Blindspot: False dichotomy (with a side of survivorship bias). The hosts let “AI forces black-and-white adjudication; humans tolerated nuance” stand as the frame, using a provider who preferred the human prior-auth reviewer — but he preferred it because it approved his care. That’s a self-interested sample, not evidence the human era was more accurate. Human reviewers were inconsistent too; they just erred in his favor.
🎙️ The Heart of Healthcare — “How AI Will Finally Make Healthcare Deflationary” (with Eric Larsen)
The thesis worth sitting with — call it healthcare’s Oppenheimer moment: “she who assumes the liability wins.” With model interpretability still unsolved, Larsen argues the market goes to whoever has the balance sheet to shoulder the malpractice and product liability — which favors hyperscalers over pure-play startups.
💡 Builder take: If liability is the moat, the clinician-builder’s edge isn’t the model — it’s owning the narrow, well-bounded decision where you can credibly stand behind the output. Scope down until the liability is something a human expert can actually hold.
💬 Standout Quote:
“We can only move at the speed of blame allocation.” — Eric Larsen
💡 BTW: The firm that just wrote Prosper its $30M check, a16z, was co-founded by Marc Andreessen — who built Mosaic, the browser that put the web on the map, as an undergrad earning $6.85 an hour in a university supercomputing-center basement. The check-writer started as the kid doing the unglamorous build for hourly wages.
📅 Upcoming: Adoption of AI in Clinical Care — HHS RFI readout (ONC/HHS), Thu Jun 25, 3 PM ET — the policy signals you’ll be designing around.
What are you building this week? Email and tell me (kevin@clinicians.build) — I read every one.
— Kevin