Deploying AI in Healthcare

Episode Summary

Executive Summary: The episode centers on how AI is rapidly reshaping healthcare, especially clinical documentation, workflow automation, and revenue cycle management. Nikhil Buduma explains Ambience’s thesis: build deep infrastructure on top of messy healthcare data, earn clinician adoption in high-complexity health systems, and use AI to unlock margin, reduce burnout, and eventually support more autonomous virtual care teams.

Main Topics: Why healthcare is ready for AI now (Priority: 5/5): Speakers argue rising demand, clinician shortages, and burnout have created urgent pressure to use AI to do more with less, and recent model improvements have finally made the technology useful in real clinical settings. Nikhil Budama’s path into healthcare AI (Priority: 4/5): Nikhil recounts leaving an MD-PhD path after a mentor died from a medical error, then working in early deep learning circles before deciding to apply AI to healthcare systemically rather than through a traditional clinical career. Why Ambience started as a care-delivery operator (Priority: 5/5): The founders ran a medical practice and implemented EHRs to gain first-hand empathy for operators, learn workflow constraints, and build the intuition needed for a later platform company. Enterprise healthcare adoption and market segmentation (Priority: 4/5): Budama argues that high-complexity academic medical centers are the hardest but most defensible market, while mid-market practices are more fragmented and competitive, with lower barriers and more vendor churn. AI as infrastructure, not just intelligence (Priority: 5/5): A major theme is that healthcare AI requires solving data extraction, decision-trace capture, quality definition, and workflow integration; model capability alone is insufficient without a strong data and product layer. Margin expansion, ROI, and the future of the EHR stack (Priority: 5/5): The discussion emphasizes that AI must prove hard ROI—through coding, throughput, denial reduction, and labor savings—to change health system economics and potentially shift power away from legacy EHRs. The path from copilot to more autonomous care (Priority: 4/5): Both speakers explore how AI could expand from documentation assistance to proactive pre-visit prep, post-visit follow-up, and virtual care team functions, though full autonomy remains constrained by safety and workflow complexity.

Key Arguments: Healthcare demand is rising faster than clinician supply, creating a structural need for AI-enabled productivity gains. Doctors and nurses will adopt technology only when it meaningfully improves their daily work; consumer-grade expectations have finally begun to translate into healthcare. Running a care delivery organization first gave Ambience the empathy and operational context needed to build a platform that works in real hospitals. Enterprise health systems are the most difficult segment, but if a product achieves high adoption there, it becomes extremely defensible. Healthcare AI is constrained less by model capability alone than by messy data, missing decision traces, and hard-to-define quality standards. Legacy EHRs are not built for AI-era product speed; companies like Ambience need an abstraction layer on top of the EHR to accelerate new product development. The future value of AI in healthcare will depend on operating-margin improvement, not just labor relief or novelty. A shared source of truth between providers and payers could reduce the value of adversarial RCM and payment-integrity battles over time. The most promising near-term autonomy is not a fully autonomous doctor but a virtual care team that can handle pre-visit, post-visit, and coordination tasks. AI-native internal workflows can dramatically increase engineering leverage, reducing the number of people needed while increasing output quality.

Data Points: People aging into Medicare per day: 10,000 - Used to illustrate the speed of healthcare demand growth and system strain. Clinician daily product usage at customer sites: 75%+ - Ambience claims high daily usage among clinicians at large academic medical centers. Visits using Ambience at customer sites: 80%+ - Budama says clinicians use the product for most visits in active deployments. Net new margin projected for one health system: $30 million - Example of ROI projected from Ambience adoption at a health system. Usage rate at some organizations with other AI tools: 15-20% - Budama says some competing rollouts see low clinician adoption. Visit coverage for some competing tools: 20-40% of visits - Examples of limited utilization even among doctors who do adopt AI tools. Model scale in earlier era: Tens of millions to ~100 million parameters - Budama contrasts early transformer-era models with today's trillion-plus models. Current model scale: Trillion-plus parameters - Used to show how much foundation model capability has expanded. Deployment cycle to live learning loops: <30 days - Ambience says it can move from concept to production learning with marquee health systems in under a month. Products on roadmap: 2 to 12 to 24 - Illustrates how Ambience's infrastructure layer increases product clock speed.

Pivotal Quotes: "We live in a world where the demand for healthcare is just rising so quickly." — Nikhil Budama: Opening argument about why AI is urgently needed in healthcare. "We're building in a world where the floor is lava." — Nikhil Budama: Describing how fast AI capabilities evolve and why healthcare AI companies must continually adapt. "There is a pathway to doing more with less." — Nikhil Budama: Summarizing the optimistic thesis that AI can reduce burden for clinicians and improve patient experience.

Implications: Healthcare AI winners will likely be those that combine strong workflow adoption, deep data infrastructure, and measurable economic ROI. The next wave may expand from copilots to more autonomous care coordination and reshape the EHR stack.