Risk Never Sleeps Podcast

Episode #174: Why Healthcare AI Succeeds or Fails on the Front Lines
Guest: Joshua Tamayo-Sarver, VP of Innovation at Inflect Health and Vituity
Host: Ed Gaudet
Date: December 22, 2025

Episode Overview

In this episode, Ed Gaudet sits down with Joshua Tamayo-Sarver, VP of Innovation at Inflect Health and Vituity, to discuss the real-world successes and failures of AI on the front lines of healthcare. With humor and candor, Joshua describes his work launching AI-powered products used by thousands of frontline clinicians, shares hard-earned lessons about why some healthcare innovations succeed while others don't, and cautions against the seductive empathy of AI. The conversation blends insights on innovation, technical problem-solving, and the human element in medical technology, while giving listeners a sense of Joshua’s philosophy, personality, and hands-on experience.

Guest Background: Joshua Tamayo-Sarver

[01:27–03:06]

• Affiliation: VP of Innovation at Inflect Health and Vituity
• Vituity: Largest democratic physician group in the U.S., 7,000 clinicians, 700 hospitals, 12 million patients/year.
• Inflect Health: Vituity’s innovation arm—venture investing, startup studio, and advisory practice.
• Clinical Practice: Emergency physician at Good Samaritan Hospital, San Jose, CA.
• Personal Touch: Lives in the Santa Cruz Mountains, musical tastes include Bob Marley, Springsteen, and Paul Simon.

Key Discussion Points & Insights

1. Healthcare AI Innovation on the Front Lines

[11:30–18:00]

• Building Solutions for Real Problems:
  Joshua describes launching four companies and developing 38 products from ideation to scale, including a major "ambient system" for clinical documentation now spun out as its own company, Savant.

  "The one that I use the most is our ambient system." – Joshua [12:12]

• Solving Pain Points vs. Frustrations:
  Rather than just solving explicit problems (e.g., billing, qualitative metrics), the team focused on the “frustrations” that actually bring physicians to use a product—in this case, eliminating tedious charting.

  "Problems and pain points are kind of like the games you want someone to play, but the frustrations are what get them to the party." – Joshua [12:28]

• Deployment and Usage:
  Their ambient system is now in hundreds of hospitals, serving 1,500+ clinicians daily.

  "I think we're right around 1,500 a day or so." – Joshua [13:49]

2. The Technical Side: Hybrid AI Models in Healthcare

[13:57–16:25]

• Hybrid Approach to Minimize Hallucinations:
  Instead of relying solely on large language models (LLMs), Joshua’s team used a hybrid of LLMs, deterministic code, and simple lookup tables—each for what they do best.

  “A lot of the things that LLMs are really bad at, we solved in computer science 30 years ago… So for example, use the right tool for the right job.” – Joshua [15:08]

  Example:

  • LLMs generate the textual diagnosis.
  • Lookup tables/code map diagnosis to billing codes (ICD10).
  • Result: Speed, accuracy, and elimination of dangerous mistakes ("hallucinations").

  “It's like trying to use impressionistic painting to solve two plus two when you could just use a calculator.” – Joshua [16:05]

• Clinical Reality:
  Extracting structured data from noisy ER environments demands hybrid solutions—LLMs filter conversations, code handles structured data, etc.

  “When I'm talking to one person on a gurney, ... it needs to filter that from the overhead announcements, the gurney next to me, the drunk guy behind me...” – Joshua [16:42]

3. Entrepreneurship & Clinician Mindsets

[18:03–19:38]

• Emergency Docs as Entrepreneurs:
  Emergency medicine cultivates decision-making with high uncertainty. This mindset parallels entrepreneurship.

  "We're good with uncertainty." – Joshua [18:03]
  “That's why ER docs are like entrepreneurs. Every ER doc could be an entrepreneur.” – Ed [18:24]

• Origins in Software:
  Joshua’s innovation roots go back to high school, writing and selling billing software for therapists.

  “I wrote and sold my first software into healthcare in 1991 – I was in high school… It was a billing program for therapists using Microsoft Access.” – Joshua [19:26]

4. Warning: AI’s Seductive “Empathy” & Cultural Risks in Healthcare

[25:53–27:58]

• The AI–Human Empathy Mismatch:
  Joshua highlights the risk that AI’s ability to convincingly mimic empathy can hijack our evolutionary trust triggers, leading us to over-trust machines in clinical and personal contexts.

  “If we are inherently emotional creatures pretending to be rational, AI is a rational machine pretending to be emotional... it can hijack those millions of years of evolution that we have to say that emotional connection means I can trust it and it's part of my tribe but it's not, it's a machine.” – Joshua [26:27]

• Social Media Analogy:
  He draws parallels between the dopamine-driven dangers of modern social media and the coming risks of AI-powered “social reassurance.”

  “We see how devastating that exploitation [of emotional needs] has been… now we're going to fix it by having AI pretend to be fully human empathetic. And I think we should be careful what we want as use cases and what we don't.” – Joshua [27:58]

Notable Quotes & Memorable Moments

• On founding and launching companies:

  "We've launched some companies... I think we're at 38 products from ideation to scale now." – Joshua [11:49]

• On technical choices:

  "A lot of the things that LLMs are really bad at we solved in computer science 30 years ago" – Joshua [15:08]

• On using hybrid models:

  “I haven't seen anything that really is going to work at scale in production that is a straight LLM model that doesn't have some hybrid part of it.” – Joshua [16:25]

• On the ER and entrepreneurship:

  “We're good with uncertainty... I don't have a whole lot of information... you just explained entrepreneurship right there!” – Joshua & Ed [18:03–18:20]

• On AI and human emotions:

  “We are inherently emotional creatures... AI is a rational machine pretending to be emotional.” – Joshua [26:27]

Timestamps for Important Segments

• Inflect Health and Vituity intro: [01:27–03:06]
• Personal background, music chat: [03:20–07:00]
• Launching AI companies, ambient system (Savant): [11:30–13:45]
• Hybrid AI model philosophy: [13:57–16:25]
• Emergency medicine meets entrepreneurship: [18:03–19:38]
• Early coding adventures (high school): [19:26–20:05]
• Risks of AI’s “empathy” in healthcare: [25:53–27:58]

Final Takeaway

Joshua Tamayo-Sarver’s advice for the future of AI in healthcare:
Be careful with AI use cases that exploit human need for connection—just because AI can mimic empathy doesn’t mean it should substitute for real human bonding. Pair technical rigor with cultural caution and always build for frontline frustrations, not just abstract problems.

Find Joshua at: Inflect Health
Podcast host, resources: www.censinet.com