Sully.ai integrates with 28 EHRs by name. Assort Health publicly claims integration with 80+ EHR systems. None of those EHRs see a dollar of the $1,500 to $10,000 per clinic per month that these AI startups bill. The incumbents who figure out the partner economics first will capture disproportionate value. The ones who treat it as a security problem will lose the AI layer above them, and eventually the customer relationship below them.
Here’s what I mean.
Over the past eighteen months, a cohort of well-funded AI startups — Assort Health ($102M raised), Sully.ai (400+ organizations), Insight Health, Paratus (YC W25), Vocca (Europe-first), Elise AI, Talkie.ai, and dozens of smaller players — have built voice AI front-office products for ambulatory and dental practices. They handle scheduling, intake, insurance verification, reminders, and payment collection. Patients call; the AI answers.
None of these products work without connecting to the EHR. And the EHRs almost never publish the APIs the startups actually need. So the startups integrate the only way they can: the clinic provides a login credential, the startup’s bot logs in as if it were staff, and the backend workflow runs on top. This happens at scale. Assort publicly claims integration with 80+ EHR systems. Sully’s own integrations page lists 28 EHRs by name including Epic, Oracle Cerner, Meditech, Altera, Veradigm, Allscripts, NextGen, AdvancedMD, Kareo/Tebra, ModMed, DrChrono, Elation, Practice Fusion, and more. Insight Health names similar coverage and explicitly markets itself as an Athena Marketplace, Epic Toolbox, NextGen Marketplace, and Charm Marketplace partner.
The economic shape of this is worth sitting with. The clinic pays the AI startup $1,500 to $10,000 per month. The AI startup runs its product on the EHR’s rails. The EHR sees none of that revenue.
Multiplied across the thousands of clinics adopting voice AI in 2025, the revenue the ambulatory EHR category is leaving on the table is not trivial. And the strategic risk compounds over time: as these startups accumulate structured longitudinal data across their customer base, the switching cost to replace the EHR underneath eventually decreases. The AI layer becomes the durable customer relationship. The EHR becomes the commodity.
Athena’s Answer (And What They Got Right)
Athenahealth recognized this dynamic earlier than most. Their response is public and worth studying. The athenahealth Marketplace now lists over 500 technology partners across 50 digital health capabilities, including a dedicated Agentic AI category launched in mid-2025. Assort Health, SOAP Health, HealthTalk A.I., Insight Health, and Voiceoc all participate. Athena’s position is explicit: rather than treating AI startups as a security threat, they’ve structured the integration path as a partner program with revenue share mechanics and co-marketing. athenaOne’s CPO Paul Brient has said publicly that this lets customers “take advantage of all this stuff without having to have an IT shop that has to evaluate all kinds of technologies.”
Epic responded similarly with App Orchard. Oracle Cerner has its Open Developer Experience program. These are the three that have figured out the playbook.
Athena got the structural answer right, but there are choices the next wave of EHRs can make differently. The 500-partner Marketplace is broad and generous to partners; tighter integration certification and a more selective partner roster would have captured more upside per partner. The lesson for Tier-2 EHRs designing their response now isn’t to copy Athena, it’s to learn from where Athena left margin on the table.
The interesting question is what the rest of the category does.
What’s Different About Tier-2 Ambulatory and Specialty EHRs
NextGen, AdvancedMD, ModMed, eClinicalWorks, Greenway, Tebra, Veradigm, Elation, and the specialty EHRs — these are where the AI startups publicly claim integrations but the partner economics are less developed. Some have marketplaces in name but not in substance. Some have none. Meanwhile, recent transactions in the space make clear that the capital and mandate are there: Thoma Bravo and MDP at NextGen ($1.8B take-private, expanded 2025), Francisco Partners reacquiring AdvancedMD ($1.1B in Dec 2024), Clearlake at ModMed ($5.3B valuation in March 2025), Hildred at Tebra ($250M raise in Dec 2025 explicitly for AI). Every one of these sponsors is publicly committed to AI-strategy acceleration.
In dental, the pattern is cleaner. Every dental AI startup I examined — Newton, Savvy Agents, ConvoCore, GetHelpdesk.AI, OpenMic, Kickcall, DentalAI Assist, AI Receptionist Dental — names the same three PMS vendors as integration targets: Dentrix, Eaglesoft, and Open Dental. Henry Schein One announced an AWS partnership in November 2025 for agentic AI across Dentrix and Dentrix Ascend. Patterson Dental’s public posture on Eaglesoft’s AI is visibly quieter. Open Dental, independent and founder-led, has no public AI-partner strategy at all.
The Partner-vs-Build Question
For EHRs deciding what to do in the next 12 to 24 months, the framing that matters isn’t “should we block scrapers” — blocking them breaks clinic workflows and the clinic blames the EHR, not the scraper. The framing that matters is three-way:
Native AI (build everything in-house). Works if you have the engineering scale and the customer concentration to amortize specialty-specific AI investment. Athena can afford this. Epic does this. For mid-tier EHRs, native-everything is capital-intensive and slow.
Partner program (structured). Works if you have enough scale to make the partner economics attractive (revenue share, co-marketing, certified integration path) and enough brand to prevent partners from disintermediating you. Athena’s 500+ partner marketplace is the reference model.
Hybrid. Native for the AI capabilities that define the EHR’s value prop; partner program for the ones that extend it. This is where most Tier-2 ambulatory EHRs should end up, but getting the cut line right — which AI is native, which is partner — is a nontrivial strategic question that should be done at the specialty or workflow level, not the EHR level.
The EHRs that get this wrong either over-build (wasting capital on AI capabilities a partner would provide better and cheaper) or over-partner (ceding customer relationship to the AI layer). The ones that get it right use partner economics to capture revenue they’d otherwise leave on the table and preserve their position as the system of record.
My read on where the cut line should sit: native for the AI capabilities that touch clinical decisioning or are specialty-deep (AI scribes tuned to dermatology, AI coding optimization for orthopedics, predictive denial prevention by specialty). Partner for the AI capabilities that are horizontal across specialties (front-office voice reception, no-show prediction, generic scheduling intelligence, payment automation). The native ones build category moat. The partner ones drive distribution revenue.
Get this cut wrong in either direction and you either burn capital or lose ground. Get it right and you turn the AI receptionist threat into a revenue line.
What I Notice
I spent most of the last decade building production systems in FinTech lending, and last year I led engineering at a legal tech company shipping three AI products on regulated data — multi-agent orchestration, RAG infrastructure, and an evaluation framework. The dynamic I’m describing in ambulatory healthcare looks very similar to patterns I watched play out in other regulated SaaS categories. The incumbents who treated the AI layer above them as a threat tried to build walls and lost ground. The ones who treated it as a new distribution channel and structured the economics accordingly captured most of the upside.
If you’re running product or strategy at an ambulatory or dental EHR and these questions are scoped for the next two quarters, I’d want 30 minutes. No pitch, no deck. I’ve seen the shape of this before in other regulated SaaS categories and am mapping how it plays out here.