Julie Yu, A16Z general partner, interviews Nikhil Buduma, CEO and co-founder of Ambience Healthcare, a company founded in 2020 that works with major academic medical centers to deploy AI-powered clinical tools.

Buduma's journey began as an MD PhD student at Stanford who dropped out after losing a mentor to medical error, then spent over a decade in the AI research community including time with researchers who founded OpenAI. He witnessed the emergence of Transformers in 2017 and scaling laws, writing Fundamentals of Deep Learning during his early career.

Rather than building technology first, Buduma and co-founder Mike started a medical practice to understand healthcare operations firsthand - implementing EHRs, working with doctors, and experiencing the challenges of running a care delivery business.

The conversation explores AI adoption in healthcare, the technical challenges of clinical intelligence, market segmentation between enterprise and mid-market providers, and the future of autonomous medical AI as demand outstrips physician supply.

From OpenAI Researcher to Healthcare Entrepreneur

Buduma started as an MD PhD student but dropped out after losing a mentor to medical error, choosing to work on systemic healthcare problems rather than becoming another clinician.

He spent time in Greg Brockman's apartment with early AI researchers who became OpenAI, working on deep belief nets and GPU scaling around 2010 when Andrew Ng was at Stanford.

During this period, Buduma wrote Fundamentals of Deep Learning, documenting the rapidly evolving field where entire branches would 'collapse' every six months as new architectures emerged.

The 2017 Transformer architecture changed everything - 'we just saw the entire research community sort of just collapse on this architecture because it was so clear that it solved many of the challenges around language modeling and reasoning' - Nikhil

Rather than build technology first, Buduma and co-founder Mike started an actual medical practice to understand healthcare operations, implementing EHRs and working directly with physicians.

Healthcare AI Adoption Reaches Tipping Point

Healthcare providers have become the fastest adopters of AI, reversing their historical position as technology laggards compared to payers and life sciences companies.

'The delta between the magic of the tool that they're experiencing in their consumer lives and what they do in their work has, for the first time, narrowed just even a little bit' - Nikhil

Ambience achieves 75%+ daily adoption rates among clinicians at major academic medical centers, using the platform for 80%+ of visits versus 15-20% adoption for competitors.

The market bifurcates between high-complexity academic medical centers requiring sophisticated AI and simpler mid-market practices where EHRs and basic AI scribes compete.

'When we even hint that we're about to release a new product, the energy we feel from the clinicians is almost like lining up for the next new Apple product' - Nikhil

The Technical Challenge of Clinical Intelligence

'AI clock speed is fundamentally different from product clock speed' - companies must build for capabilities 18 months ahead while the 'floor is lava' with rapidly evolving AI capabilities.

Most EHRs use mutable data structures that destroy decision traces, requiring complete architectural rethinking to capture the clinical reasoning process for AI training.

Data integration remains unsolved - pulling context from FHIR APIs and proprietary systems with inconsistent data models where 'people just stuff like free text into a random field'.

Defining quality in healthcare AI is 'fundamentally complicated' - two doctors agree on ICD-10 coding only 60% of the time, and adding a coder creates three different answers.

Foundation models lack healthcare-specific context like chronicity, credentialing hierarchy, and the gap between spoken words to patients versus clinical documentation requirements.

Revenue Cycle and Operating Margin Transformation

One health system projects over $30 million in net new margin from Ambience through improved revenue cycle management, throughput, and reduced need for human scribes.

Early AI scribe adoption focused on physician retention and happiness, but has shifted to 'hard ROI' with measurable financial impact for CFO offices.

Ambience tracks user behavior inside EHRs, maps to specific codes, prevents CDI queries and denials, then attributes to new cash collection - requiring downloading entire data warehouses.

'This is the first time that we as an industry, that there is a class of technologies that can fundamentally change operating margin' - creating compound growth flywheels.

The provider-payer AI arms race may resolve positively - shared source of truth could make both RCM and payment integrity teams economically inefficient within five years.

The Path to Autonomous Medical AI

'We have 10,000 people aging into Medicare every single day, and we just can't train doctors fast enough to take care of all these people' - driving need for AI augmentation.

Ambience experiments with pre-visit AI agents that anticipate clinician needs, review patient data for hours, and prepare comprehensive summaries before appointments.

Post-visit AI agents could provide continuous patient engagement - answering questions, checking medication pickup, ensuring lab completion, and managing anxiety around procedures.

Current bottlenecks include cascading context across care settings and predictive modeling, though 'so much is buildable' that capability limits aren't the primary constraint.

'There is a pathway to doing more with less. There is a pathway for the job of being a clinician, being a nurse to be a fulfilling one' - Nikhil

Building AI-Native Healthcare Companies

Individual engineers accomplish dramatically more with tools like Claude Opus 4.5, requiring 'really smart thinkers' but fewer people overall to complete substantial work.

Hiring profiles have shifted toward platform engineers who think about long-term architecture and product engineers who embed in clinical environments for requirements gathering.

Internal AI tools help new employees access 'decision traces' and historical context, enabling them to 'stand on the shoulders of giants' when joining the organization.

'If you were building a company today, you would not do it the way you would be doing it two years ago' - fundamental changes in organizational structure and capabilities.

Ambience built a data layer above EHRs that dramatically reduces incremental cost of new AI use cases, enabling expansion from 2 to 12 to 24 products with shared infrastructure.