AI-Enabled Care: What’s Actually Different Now

Telehealth 1.0 optimized the visit. What comes next should be measured by what those visits open up.

When I’m sick, I still turn to general-purpose AI, and think most readers of this piece will likely do the same. The question is no longer whether or not Americans will use AI for their healthcare. The question is whether we can build a business that is substantial enough to become a patient’s real primary care relationship and financially sustainable enough to survive the realities of operation that have challenged every attempt at this space so far.

The hard constraints

A lot of what happens in a visit still needs a physician: ambiguous symptoms, messy test results, medication tradeoffs, and the trust patients need when decisions feel high stakes. Prescriptions and lab orders still require licensed physicians prescriptions and requisitions by law. Earlier this year, Utah’s AI Learning Lab started to allow for select Rx refills through an AI system; other states may follow, but broad authority around Rx/labs will take much longer to be given to AI systems. As such, a platform must have doctors on staff (to validate system guardrails and provide patient care) if it wants to be clinically useful and differentiated from general-purpose models… and that’s when the real challenge begins.

Licensing. When a doctor treats someone, they need a license in that patient’s state. Getting one provider licensed across a meaningful number of states runs ~$50k. Lose that doctor, and suddenly your coverage in those places vanishes overnight. Finding a 50-state licensed physician is nearly impossible, particularly ones that perfectly match companies’ desired staffing hours.

Staffing (against seasonality). W-2 physicians provide consistency in scheduling and patient experience, but have lock in cost. Per-diem physicians who get paid per visit give you the flexibility, but can create quality or availability challenges. High demand seasons stir constant choices - push extra cases to your full-timers if they’ve got room, or lean on per diems who need that visit volume to stay engaged? Summer slows things down. Fewer patients show up, and physician efficiency drops, though schedules stay full (and you need to make sure you are still giving enough volume to per diems who get compensated per visit). Margins might actually tick up a little, but you cannot plan around that because flu season and the holiday surge are always coming. This never really stabilizes.

Panels vs. pooling. If you want patients assigned to specific doctors for continuity, those doctors stop being interchangeable. You keep the relationship but lose pooled scheduling efficiency and the ability to meaningfully tap into your per diem physicians, because each doctor’s calendar is now dictated by panel, not overall demand.

Pricing. Patients reach telehealth through three paths: DTC apps, their employers, or bundled into the payor’s benefit stack (We’re setting aside B2B motions that power other digital health companies; that’s a different discussion). DTC could work, but CAC is high and will likely draw in patients that utilize physician-in-the-loop services more than you had modeled for. Companies try to cut costs by providing physicians the intake and workflow layer to reduce visit length to a “click”. However, supply-demand matching has lagged behind workflow efficiency gains, and savings have not caught up to buyers’ demand for more affordable rates that buyers demand. So the math got tighter. Everyone ended up stuck optimizing cost and revenue per visit as if those were the only numbers that mattered.

An obvious tension against unit economics here is latency/wait time. While simple Rx-related visits can truly be reduced to a click, for the more involved visits, it’s critical to figure out how to retain the differentiation from the large language models or Rx-focused telehealth wrappers, in figuring out the balance here to ensure you aren’t sacrificing access but is able to make the economics work. We’d love to hear from builders who are finding that balance.

Beyond the visit

Let’s assume that cost (in the above scenario) is mostly fixed. Revenue per visit, if you are only doing telehealth, is mostly capped. So the real question is: how do you create value that goes beyond the visit? What do the patients want and need?

The provider who remembers everything. Today, each appointment means telling everything again - past visits, medications, how things have felt. Some platforms are starting to change this, building systems that save details between meetings: history, drug names, diagnoses, preferences (and deliberately making the system visible to patients). Yet the memory should stretch outside one platform. If a person sees a specialist, results must flow back somehow. We have made and seen promises of interoperability for over a decade and have mostly failed to deliver. But the current environment gives us some hope (i.e., TEFCA is live with over 10,000 participating organizations, Epic Nexus is bringing ~280M patient records on, HHS escalated information blocking enforcement to $1M per violation in 2025, and the list goes on). Far from solved, but companies are building here, and the tailwinds are a bit more concrete than they’ve been.

Next Best Action Engine. When a system starts to hold your past visits, blood tests, doctor notes - it should say something real. Book this screening. Revisit that dosage given your last labs. Follow up on that result from three weeks ago. Once wearable data, CGM readings, and imaging start flowing in (not there broadly yet, but coming), the recommendations begin to fit tighter. When it makes sense to see a specialist, the platform should find someone in-network and taking new patients, book the visit, share the relevant history, and check in afterward to make sure prescriptions got picked up and things went smoothly. This builds trust and habits.

Patients’ Investigation Partner. Not every patient intent means something sudden. Many wait weeks between specialist appointments, handle medication side effects, or try to untangle a new diagnosis. Right now, these people open ChatGPT before calling the clinic. They’re not chasing another visit. What matters is clear insight into their own case - something personal, grounded in their actual medical history, that shows them how things stand and what to ask about next. Right now, nobody pulls this off well (think you can technically use OpenEvidence a few times without an NPI). Whoever cracks it first builds a platform patients choose to come back to, not only out of habit, but also as it holds their full medical history, and comes with on-demand clinicians who can step in when questions fall into gray areas.

The visit has been the entire product for too long. We think the real value resides beyond the visit. For the first time, the regulatory infrastructure and the technology are both moving in the right direction to make this work. We’re excited to meet the founders and operators building here.