10 Million Americans’ Healthcare Behind the Scenes
10 Million Americans’ Healthcare Behind the Scenes
Show Notes
Eugene Sayan is the founder and CEO of Softheon, a 25-year-old bootstrapped startup that processes $20 billion in premiums, powers 30 million enrollments, and delivers affordable healthcare access to 10 million Americans. Named to Time Magazine's World's Best Tech Companies of 2025, Softheon operates as the white-label infrastructure layer behind hundreds of ACA and Medicare health plans - running 1,300 agentic processes per average customer with a target accuracy of 99.9999999%. Eugene filed a patent in 1998 for agentic AI frameworks, predating the current wave by more than two decades.
This conversation goes deep on what zero hallucination tolerance actually means in practice, how to build a company that survives 25 years of political headwinds, why the Infinite Game is the only viable strategy for regulated industries, and how Eugene's immigrant curiosity and paranoia became his two greatest entrepreneurial superpowers.
Zero Hallucination Tolerance - What It Actually Takes
Eugene uses the US air traffic control system as his benchmark. 17,000 commercial flights per day. A 99% accuracy rate means 170 crashes. That's not a rounding error - it's a catastrophe. Healthcare operates at comparable stakes: medication dosing, eligibility determinations, enrollment processing, compliance documentation for 10 million people. Softheon's target is 99.9999999% accuracy - better than air traffic control.
Achieving that means AI cannot operate as a black box that occasionally hallucinates and gets corrected. Every agent process must have traceability (you can follow every decision back to its input), auditability (an external observer can verify the decision was correct), and human-in-the-loop checkpoints at every point where hallucination risk is non-zero. Their internal AI agent framework, named HAL - deliberately referencing 2001: A Space Odyssey's warning about AI systems that turn on their operators when poorly managed - encodes this principle into the system architecture.
Eugene's framing of current AI maturity: ChatGPT was born for the general public in 2022. It's five or six years old - a pre-K child. Fast, capable, learning rapidly. But we're approaching the teenage years: confident, sometimes wrong, prone to making up stories. Healthcare can't afford to learn from that in production. The answer isn't to avoid AI; it's to design systems that contain AI's current limitations rather than pretending they don't exist.
The Infinite Game - Navigating 25 Years of Political Headwinds
Softheon has operated through Obama, Trump, Biden, and Trump again. ACA has been attacked, defended, modified, and survived. Every administration shift creates a headwind or tailwind for a company whose entire business model is built on government-sponsored health insurance programs. Eugene's response, distilled from Simon Sinek's Infinite Game: you can't control the wind. You adjust your sails and keep sailing.
The deeper principle: when you're playing an infinite game - no defined end state, no declared winner, no timeframe - the goal isn't to win the current round. It's to stay in the game. That reframe changes every short-term decision. You don't optimize pricing to capture the current administration's tailwind; you build the capabilities the next administration's headwind will require. You don't pick sides; you stay bipartisan, because the 10 million Americans who need affordable healthcare don't care who's in the White House.
The bootstrapped structure reinforces this. Without a VC board pushing for growth-at-all-costs to hit fund return timelines, Eugene can make 10-year decisions. He can hire graduates and keep them for 15 years. He can invest in capabilities that won't pay off for five years. The organizational independence creates the time horizon the infinite game requires.
Designing for the Problem 10 Years Out
Eugene's recurring principle: Softheon solves problems that will exist five to ten years from now, starting today. He was involved in Massachusetts' health insurance exchange under Governor Romney before it became the ACA model. He filed the agentic AI patent in 1998. The 1,300 agent processes their customers now use weren't built in response to the current AI wave - they were the foundation already in place when the wave arrived.
The practical implication: when you design for today's problem, you build a system constrained by today's technology and today's regulatory environment. When you design for the problem that will exist in a decade, you build for the architecture that technology will make possible and regulation will eventually require. Companies that designed healthcare IT for the regulatory environment of 2010 are fighting the architecture choices they made then. Softheon's architecture was designed to adapt, because it was built by someone who didn't come from healthcare and therefore didn't inherit its incumbency constraints.
The outsider advantage: Eugene came to healthcare with no prior industry experience, no mental model of how health insurance IT was supposed to work. That first-principles thinking - combined with a computer science and electrical engineering background - produced a system that established players couldn't have designed, because they were too constrained by how things had always been done.
Frameworks from This Episode
- Zero Hallucination Architecture - Designing AI systems for high-stakes environments by requiring every agent output to have traceability (decisions traceable to inputs), auditability (external verifiability), and human-in-the-loop checkpoints wherever hallucination risk is non-zero. In healthcare, air traffic, or financial compliance, a 1% error rate isn't a small problem - it's a catastrophe measured in lives and liability.
- The Infinite Game Strategy - From Simon Sinek: competing without a defined end state. No winner declared, no timeframe, no score. The goal is to stay in the game long enough that short-term political or economic headwinds become irrelevant. Requires bipartisan mission alignment, long-term hiring, and organizational independence from pressures that optimize for current-cycle performance.
- 10-Year Problem Design - Build solutions for the problem that will exist five to ten years from now, not the problem visible today. Requires outsider perspective (no inherited incumbency constraints), first-principles thinking, and organizational structures (bootstrapped, long runway) that support multi-year investment horizons. The payoff: when the future arrives, you already have the infrastructure rather than rebuilding it under competitive pressure.
Tools Mentioned
- Softheon - AI-powered healthcare technology platform. White-label infrastructure for ACA and Medicare health plans. Processes $20B in premiums, powers 30M enrollments, and serves 10M Americans. Runs 1,300 agentic processes per average customer with zero hallucination tolerance. Built on 25 years of proprietary agentic AI infrastructure.
- Oura Ring - Wearable health tracker. Eugene uses it for continuous sleep and recovery monitoring; data feeds into Apple Health and his EMR system so physician appointments begin with biomarker review rather than basic health status questions.
Glossary
- Hallucination Tolerance - The acceptable error rate for AI outputs in a given system context. In consumer chatbots: high tolerance (wrong answers are annoying, not dangerous). In healthcare, financial compliance, or air traffic control: zero tolerance (wrong outputs create liability, harm, or death). Designing for zero hallucination tolerance requires human-in-the-loop verification, transparent reasoning chains, and agent process traceability - not just better models.
- Human-in-the-Loop (HITL) - An AI workflow design pattern in which human review and approval is built into the execution path at defined decision points, rather than humans reviewing outputs only after execution completes. In high-stakes environments, HITL is not a workaround for AI limitations - it's the architecture. The human doesn't replace the AI; the human provides the accountability layer the AI cannot yet provide for itself.
- Agentic Framework - A software architecture in which autonomous digital agents handle discrete, well-defined tasks within a larger workflow - passing outputs to the next agent, escalating to human review when thresholds are crossed, and operating continuously without manual triggering. Softheon's average customer runs 1,300 such agent processes. Eugene filed a patent for this approach in 1998, predating the current AI wave by more than two decades.
- The Gartner Hype Cycle - Technology analyst firm Gartner's model for how new technologies move through public perception: from technology trigger → peak of inflated expectations → trough of disillusionment → slope of enlightenment → plateau of productivity. Eugene maps current LLM adoption to the peak of inflated expectations (AI as panacea), predicting a coming trough as limitations become apparent in production environments - followed by real, sustainable adoption as guardrails and architectures mature.
- B2B2C - A business model in which a company sells technology or infrastructure to a business customer (B2B), which in turn delivers that technology to end consumers (B2C). Softheon is the middle B: they build the platform that health insurance companies use to enroll, retain, and serve their individual members. The 10 million Americans using Softheon's technology typically don't know Softheon's name - they experience it through their insurance company's branded interface.
- Pharmacogenomics - The study of how an individual's genetic makeup affects their response to drugs - including which medications will be effective, which will be ineffective, and which may cause adverse reactions. One branch of personalized medicine. As genetic sequencing costs collapse and AI accelerates biomarker analysis, pharmacogenomics enables physicians to prescribe based on individual biology rather than population-average clinical trials. Eugene sees this as a near-term frontier for AI-assisted healthcare.
- The Infinite Game - Simon Sinek's framework (from the book of the same name) for organizational strategy: in infinite games, there is no defined end state, no declared winner, and no timeframe. Participants can join and leave. The goal is not to win but to perpetuate play - to stay in the game long enough that you outlast those playing with finite-game thinking. Applied to Softheon: every administration shift is a finite-game event. The company's mission to provide affordable healthcare access is infinite. Finite-game pressures (political headwinds, economic cycles) become manageable when the organizational identity is grounded in the infinite game.
Q&A
How does Softheon achieve 99.9999999% accuracy with AI?
Two architectural decisions that work together. First: every AI agent process has traceability and auditability built in - every output can be traced back to the input that generated it, and verified by an independent observer. Second: human-in-the-loop checkpoints are embedded in every process where hallucination risk is non-zero. The human doesn't replace the agent; the agent does the computational work and the human provides the accountability layer. Their internal agent framework is named HAL - a deliberate reference to 2001: A Space Odyssey - as a constant reminder that AI systems that are poorly managed can turn against their operators. The name is the policy.
How does Softheon navigate political headwinds when its entire business depends on government programs?
Eugene's sailing analogy: you can't control the wind or the weather. You adjust your sails and keep sailing. Practically: stay bipartisan. Don't optimize for the current administration's policy direction. The 10 million Americans using Softheon's technology don't care who's in the White House - they need medication access and care regardless of which party controls the executive branch. The bootstrapped structure supports this; without VC pressure to maximize current-cycle growth, Softheon can make decisions on a 10-year horizon that short-term political cycles can't disrupt.
Why did Eugene build Softheon without venture capital after the initial raise?
The last outside capital was raised during the dot-com bubble in 1999. Since then, organic growth only. The bootstrapped model creates organizational independence: no investor board pushing for growth-at-all-costs to hit fund return timelines. This enables long-term decisions - hiring graduates and keeping them for 15 years, investing in capabilities that won't pay off for a decade, designing systems for the problem that will exist in 10 years rather than the problem that needs to be solved this quarter. The pressure and privilege of organic growth: you learn to live within your means, which also means you never need permission to stay true to your mission.
What's the bear case for AI in healthcare?
Eugene identifies two risks. First: hallucination at scale. As LLMs get integrated into clinical decision support without adequate guardrails, errors will surface. The AI is currently a pre-K child growing rapidly - we're approaching the teenage years where it's confident and sometimes wrong. Decisions made now about accountability architecture will determine how much damage those errors cause. Second: AI as a cybersecurity backdoor. LLM models deployed without governance are black boxes - a malicious actor could distribute a model that functions as malware. Eugene draws the analogy to email: a transformative technology that became one of the most dangerous fraud vectors in history within a decade of mass adoption.
What's Eugene's vision for personalized medicine and AI?
Three converging forces in the next 5–10 years: (1) consumer-directed healthcare, where individuals engage with their health like they engage with their 401k - making informed decisions, shopping plans, managing defined-contribution health savings accounts; (2) real-time biomarker-informed care, where continuous monitoring data (wearables, CGM, sleep trackers) flows directly to physicians and enables meetings centered on data rather than basic status questions; (3) pharmacogenomics and epigenetics - personalized medication protocols based on individual genetic profiles, turning genes on and off to treat conditions like cancer that are inherently personalized diseases. Eugene is already living version one of this: Oura Ring + CGM + Apple Health + EMR integrated, all feeding his physician before every appointment.
What industry would Eugene enter if not healthcare?
Education. He cites his grandfather, an elementary school teacher, as the source of his reading habit and curiosity. His view: after more than a hundred years, the education system is structurally unchanged - students in chairs, teachers at the front, maybe a smart board instead of a chalkboard. The concept hasn't changed. AI creates the possibility of truly personalized education that adapts to how individual students actually learn, rather than forcing conformity to a single delivery model. He sees this as the next major frontier where technology could serve people at the same depth that healthcare has served Softheon's 10 million Americans.