Summary
The episode explores how AI—particularly foundation models and downstream agentic systems—can reshape clinical workflows to improve clinician efficiency, patient experience, and hospital margins. Nikhil Buduma explains why Ambience began by operating a medical practice to learn real-world EHR, workflow, and financial pain points before building a platform. The conversation highlights that the hardest work is integration: extracting high-fidelity context from heterogeneous EHRs, preserving decision traces, and defining quality for open-ended clinical tasks. They discuss market dynamics (many entrants in mid-market; defensible moat at large academic centers), the necessity of measurable ROI for CFO adoption, and the ongoing debate over generalist versus domain-specific models and the proper human/AI balance in clinical decision-making.
Key Takeaways
- 1Start with operations to build product that truly fits clinical workflows.
- 2Integration and data engineering matter more than model size for clinical impact.
- 3High clinician adoption plus measurable financial ROI unlocks enterprise deals.
- 4Market will bifurcate: crowded mid-market vs. defensible enterprise moat.
- 5The role of foundation models in clinical AI remains an open tension between generalist and domain-specific approaches.
- 6Keep clinicians in the loop; fully autonomous AI doctors are not the immediate path.
Notable Quotes
"We have 10,000 people aging in the Medicare every single day."
"Over 75% of clinicians use the product daily, and one health system is projecting $30 million in net new margins in the platform."
"We're building a world where the floor is lava."
"There's still a massive last mile problem for these models to be effective in health care."
Episode questions
Why did Ambience start by running a medical practice before building a platform?
They ran a care delivery asset to deeply understand operator pain points (EHR integration, clinician workflows, finances) and to iterate quickly; that operational experience informed platform design and change-management practices required for enterprise adoption.
What are the main technical barriers to applying foundation models in clinical settings?
Key barriers include extracting high-fidelity context from heterogeneous EHR instances and proprietary APIs, preserving decision traces in readable data structures, and defining 'quality' for open-ended clinical tasks; solving these requires substantial engineering and data-layer work beyond base model training.
How does Ambience demonstrate financial ROI to hospital CFOs?
They track user behavior inside the EHR, connect actions to downstream events (coding, CDI queries, denials), and build analytics from the data warehouse to attribute prevented costs and throughput gains—enabling CFO-level proof points like projected net new margin (e.g., $30M).
Where will most competition occur in the clinical AI market?
Mid-market and small practices will see many entrants and EHRs aiming to be AI-first, but enterprise/academic medical centers are high-complexity spaces where few vendors can meet requirements—making that segment both harder to serve and more defensible if done well.