Your AI Doctor Is Coming | Julie Yoo

1. Key Themes

Healthcare's Sunk-Cost-Free Leapfrog Advantage

Because healthcare historically underinvested in enterprise software (unlike other industries that spent decades building SaaS/workflow "middleware"), it can skip straight to agentic AI rather than ripping out legacy systems. This is framed as an unexpected structural advantage of being a laggard industry.

"We only had really like the ERP layer with EHRs and then labor. Like we were just throwing bodies at every problem. And so in some ways, we have less of a sunk cost bias as an industry... We can actually just go directly to this agentic form of technology." [00:14:11]

The First Organic (Non-Forced) Adoption Wave in Health Tech

Every prior tech adoption cycle in healthcare was externally coerced (government payments for EHRs, COVID forcing telehealth). AI scribes represent the first time healthcare workers are adopting technology purely because it's good, creating a genuine PLG dynamic.

"This is currently is the only, like the first real organic adoption wave that we've seen in health tech where doctors are just using AI scribes because they freaking work and they're so good and they really change the nature of their job for the positive." [00:15:10]

Consumers as the New Payer

A structural shift from designing healthcare products for insurers/doctors toward designing directly for consumers, enabled by AI collapsing the cost of delivering credible medical services.

"You're seeing a whole proliferation of an entirely new marketplace of products that were designed for the consumer, not like historically where you had to design for the insurance company and the doctor and then consumers almost as an afterthought. Consumers are now at the top of that pile." [00:21:27]

AI-Native and AI-Proof as the Winning Combination

The best new healthcare companies pair AI-native economics with defensible, hard-to-replicate real-world execution (in-person care, diagnostics, regulated services) — mirroring "full-stack challenger" companies from earlier tech eras.

"The best companies right now are those that are AI native and AI proof... on the inside they were obviously highly AI native and therefore had a surface area that was not achievable through historical means and a cost structure that was disruptive such that they can continue to invest in innovation because they had high margins." [00:21:56]

Healthcare's Dysfunction Is a Designed Outcome, Not an Accident

The system isn't "broken" by chance — the third-party payer model and misaligned incentives produce exactly the outcomes we see, which is why some founders are attempting to rebuild the payment system from scratch rather than patch around it.

"It's actually designing exactly as planned. Like, if you actually study the innards of how incentives are aligned based on the payment system that exists in healthcare, it results in exactly what we experience today." [00:23:25]

The Missing Data Problem for Medical AI

Because patients only interact with the healthcare system sporadically (e.g., once a year), the data underlying today's medical AI is incomplete — there's no continuous longitudinal record of a person's health, meaning current models are trained on fundamentally sparse data.

"There's a whole narrative arc of your experience as a patient that is absent from pretty much every data set that exists today." [00:25:26]

From Transactional AI Advice to Lifelong AI Doctors

Current consumer use of AI for health (ChatGPT, Claude) is highly transactional (one-off Q&A). The next opportunity is longitudinal — an AI relationship that persists with a person over their lifetime, handling diagnosis, follow-through, and next issues.

"A lot of the way that people are using, which is great, medical AI today is highly transactional. Imagine if you can have a doctor for life that's powered by that kind of modality." [00:19:06]

Cost-Shifting to Consumers Created the Market for Disruption

The rise of high-deductible health plans (employers shifting cost burden to employees) is what made consumers price-sensitive enough to seek alternatives, directly enabling the current wave of $10/month cash-pay healthcare startups.

"We have amazing companies that you can pay literally $10 per month out of pocket to get incredible, incredible services that when compared to what you could get for that same amount of dollars with the traditional healthcare system is just a thousand times better." [00:12:06]

2. Contrarian Perspectives

High Demand Doesn't Mean Full Capacity Utilization

Conventional wisdom says patient access problems stem from doctor shortages. Yoo's own company discovered the opposite paradox while building Kairos: doctors were significantly underutilized even as patients waited months for appointments — the bottleneck was routing/inventory, not supply.

"Yes, people are waiting out the door. But if you actually look at the way that our physicians are being utilized, we're still way under 100%, which was shocking, right? We had this paradox of demand out the door." [00:03:30]

Being a Laggard Industry Is Now an Advantage, Not a Handicap

Rather than treating healthcare's historic underinvestment in technology as a permanent liability, Yoo argues it's precisely why the industry can now leapfrog directly to AI-native infrastructure while other industries are burdened by expensive legacy SaaS stacks built over the last two decades.

Regulatory Constraints on Doctor Licensing Were "Deliberately" Designed to Constrain Supply

Yoo frames state-based medical licensing not as a neutral safety measure but as an intentional supply constraint that COVID forced regulators to dismantle — implying the pre-COVID system was artificially scarce by design.

"We sort of deliberately constrained supply in healthcare in weird ways. So you have to get licensed at the state level. And it was not the case that doctors typically would have a license to practice medicine outside of their state." [00:06:20]

Consumer Health Was Uninvestable Seven Years Ago — Now It's the Hottest Category

A near-total reversal in institutional judgment: Yoo says pitching a consumer health startup to VCs seven years ago would get you thrown out of the room for lacking a viable business model, yet it's now framed as the single biggest opportunity area in health tech.

"If you had come to any investor, let's call it, you know, seven years ago and pitched a consumer health business, you would have been kicked out of the room because there was no viable business model." [00:20:28]

3. Companies Identified

Kairos — Julie Yoo's own company (co-founder and CPO for nine years), built a unified system of appointment inventory across healthcare systems to route patients efficiently and solve the "patient access paradox." Sold in the past year.

"How do you sort of solve this patient access paradox by creating a single system of appointment inventory across the healthcare system and the ability to route patients to the right doctor and the right appointment." [00:03:30]

Council Health — An a16z portfolio company; an AI-native doctor's practice offering 24/7 asynchronous chat with real, licensed physicians who can prescribe, refer, and diagnose — blending AI-native UX with genuine clinical authority.

"Imagine chatting with your chat bot but having an actual licensed MD in the chat who can fully prescribe, who can fully refer, who can fully diagnose. And so you kind of get the best of both worlds of an AI-native care with real doctors who can actually provide clinical utility." [00:17:06]

Devoted Health — Cited as an example of an ambitious company rebuilding the healthcare payment/insurance system from scratch rather than working within the existing incentive structure.

"There's companies like Devoted Health and others who are doing this [building a new payment modality]." [00:23:55]

Anthropic (Claude) / AWS — Mentioned in the context of the "Rare Disease Real Kid Hackathon," which used a child's genome and clinical data (with parental consent) to generate diagnostic insights for rare pediatric diseases, illustrating AI's potential in underserved specialist areas.

"They actually had a child, they uploaded the kid's genome and the clinical data and they shared it with the community, of course, with parental consent. And people were able to actually...generate insights." [000:17:31/18:01]

OpenAI (ChatGPT) — Referenced repeatedly as a primary consumer touchpoint for health questions and as a forward-integrating layer connecting AI advice to real-world care marketplaces (e.g., booking real doctor appointments).

"They're starting to like forward integrate into actual marketplaces where you can get advice from ChatGPT and then go book an appointment with a real doctor." [00:15:39]

4. People Identified

Julie Yoo — General Partner at a16z leading healthcare investing; former software engineer turned healthcare technologist; co-founder and CPO of Kairos (sold in the past year); previously worked in genomics/computational biology. Described as a pioneer who was "too early" multiple times across her career, now positioned as one of the most credible voices on healthcare's AI transformation given her 18-year, full-cycle vantage point (builder → operator → investor).

"It was like a 15-year overnight success." [00:05:22]

Dario (Amodei, implied — Anthropic CEO) — Referenced by Sofia Puccini regarding his stated ambition to focus Anthropic's efforts on curing cancer, cited as evidence of top AI labs "doubling down" on healthcare as a frontier problem.

"Even if what you're seeing with Dario, where he's saying, okay, we're doubling down in health. We're going to cure cancer." [00:07:59]

5. Operating Insights

Find the Utilization Gap Before Building the Solution

Yoo's founding insight for Kairos came from directly interviewing healthcare executives and discovering a hidden inefficiency (doctors under-100% utilized despite visible patient demand) that wasn't apparent from the consumer-facing symptom alone. Operators should look past the visible bottleneck (patient complaints) to the underlying operational data (utilization rates) before designing a fix.

"When me and my co-founder started to talk to executives who were running these large hospitals and physician groups... what we didn't realize was they were telling us, yes, people are waiting out the door. But if you actually look at the way that our physicians are being utilized, we're still way under 100%." [00:03:01]

Price the Gap Between Alternatives, Not Just Your Own Product

The core driver of consumer adoption isn't that AI-native health products are "good" in absolute terms — it's that the delta between the existing options (four-month wait, or $2,000 ER visit) and the new option (open your phone) is so large that low friction alone wins, even without full regulatory validation.

"The trade-off is I either have to wait four months to get a real doctor's appointment or go to an emergency room and pay $2,000 or I can just open up my phone. And so I think that trade-off... makes it incredibly easy for people to just go to the lowest friction place." [00:16:09]

Build Full-Stack, Not Point-Solution, to Get AI-Proof Margins

Founders should look at historical "challenger" companies (full-stack retailers/service providers that looked ordinary on the outside but were technologically differentiated inside) as the model for building defensibility — the AI-native backend funds a cost structure competitors relying on legacy operations can't match, which in turn funds continued innovation.

"From the outside they might just look like a retailer or just look like some kind of service provider, but on the inside they were obviously highly AI native and therefore had a surface area that was not achievable through historical means." [00:22:26]

6. Overlooked Insights

The Data Substrate for Medical AI Doesn't Actually Exist Yet

Buried near the end of the conversation is arguably the most consequential structural insight in the episode: current medical AI is being trained on fundamentally incomplete data because patients only touch the formal healthcare system sporadically (roughly once a year for a checkup). This means the entire current generation of medical LLMs is missing the continuous "narrative arc" of a patient's actual health experience — implying that today's most-hyped medical AI products may hit a ceiling until new continuous-data business models (the "N of one" experiments Yoo mentions) generate the missing longitudinal datasets. This is a quiet admission that the current wave of medical AI is built on a structurally deficient foundation, not just an early but sound one.

"I would argue that we actually don't even have the data sets necessary to truly train like a medical-grade AI because we rely heavily today on electronic health record data, which is really sporadic... So there's a whole narrative arc of your experience as a patient that is absent from pretty much every data set that exists today." [00:24:57]

Regulation Didn't Just Slow Healthcare Down — It Actively Manufactured Scarcity That AI Is Now Dissolving

While Yoo frames state-level physician licensing mainly as a COVID-era anecdote, the deeper implication is that a substantial share of the "healthcare access crisis" was not caused by an actual shortage of medical expertise but by artificial jurisdictional walls preventing supply from reaching demand. This reframes the AI opportunity: much of the "unlock" isn't AI creating new intelligence, but AI-enabled business models (async chat, cash-pay, national telehealth) finally routing around regulatory scarcity that was never about true capacity limits — a much larger and more durable market opportunity than simply "AI makes doctors smarter."

"It was not the case that doctors typically would have a license to practice medicine outside of their state. But with telehealth, all of a sudden you had this national scale demand for clinical services. And so lots of states and federal agencies had to kind of rethink from first principles, how do we relax some of these constraints." [00:06:20]