# AI in Healthcare Series: Have We Already Bent the Healthcare Cost Curve?

## Executive summary

The discussion explores whether Artificial Intelligence (AI) is already beginning to bend the U.S. healthcare cost curve, suggesting that the industry is undergoing a profound, exponential shift from a model predicated on the scarcity of human expertise to one of abundance. Speakers argue that this shift—driven by technologies like Large Language Models (LLMs)—will systematically dislocate established, prestigious professions and necessitate a complete re-architecture of healthcare systems, moving away from centralized, specialized institutions toward decentralized, agentic, and first-principles designs.

## Key takeaways

- AI Performance vs. Human Expertise: While some papers suggest autonomous AI may exceed physicians in narrow, verifiable tasks (e.g., radiology), the consensus is that the comprehensive collection of tasks required for medicine is not yet benchmarked. The core challenge is reinventing the practice of medicine to leverage AI's benefits while retaining necessary human judgment.
- The Shift from Scarcity to Abundance: The current $6 trillion healthcare industry is built on the premise of cognitive scarcity (e.g., 12 years to train a radiologist). AI is democratizing expertise, making cognition abundant. This fundamental change threatens the economic models of institutions built to guard and arbitrate this scarcity.
- Historical Cost Curve Bending: Cost reduction is not solely dependent on AI. Historically, cost curves have bent due to non-AI factors, such as site-of-care shifts (inpatient to outpatient) and the introduction of new technologies (e.g., generics vs. branded drugs). This suggests that the industry has a roadmap for cost reduction even before full AI integration.

## Technical details

- AI Benchmarking and Cognitive Domains: The discussion references papers suggesting AI superiority in specific cognitive domains, noting that superimposing human judgment on top of a functionally verifiable domain can degrade AI accuracy. The focus must shift to creating comprehensive benchmarks for the entire scope of medical practice.
- Systemic Disruption and Incumbency: The analogy of Cursor vs. Microsoft (2024) illustrates how unencumbered, use-case-focused startups can rapidly outpace large incumbents with legacy systems. This suggests that the next generation of healthcare systems must be built 'in parallel' or 'from scratch' rather than attempting to retrofit intelligence onto existing, complex infrastructure.
- Agentic Systems and LLMs: The future of healthcare is predicted to involve 'agentic' interactions, where the intelligence layer operates primarily from agent-to-agent and agent-to-software, democratizing expertise and allowing individuals to become polymaths. This challenges the traditional role of the specialized academic medical center.

## Practical implications

- Healthcare organizations must plan for the systematic disintermediation of specialized human expertise by AI.
- Architectural design must prioritize building 'in parallel' or 'from scratch' to avoid being constrained by legacy systems.
- Focus should shift to developing agentic workflows and systems of record that can handle intelligence operating between autonomous agents, rather than relying on human-clicking interfaces.

## Topics

Artificial Intelligence, Healthcare Economics, Digital Transformation, System Architecture, Exponential Technology, Autonomous AI exceeding physicians paper, Cutler and Clarett Harvard paper, Man Machines in modern times

Source: https://www.youtube.com/watch?v=pUF0o57YdL8