Healthcare systems are currently trapped in "pilot purgatory," a frustrating cycle where dozens of disconnected AI tools are tested but never fully integrated into clinical workflows. Keith Figlioli (managing partner at LRVHealth) argues that the traditional software buying strategy—one tool for one department—is fundamentally broken in the age of generative AI. To achieve real return on investment, leaders must stop shopping for features and start designing a cohesive enterprise architecture.

This architectural shift relies on a three-tiered stack. At the bottom lies the core infrastructure of cloud providers and "systems of record" like Epic, which handle the heavy lifting of data and security. The middle layer consists of foundation model platforms that act as a control plane, managing how models interact with proprietary data. Finally, specialized startups sit at the top, solving specific problems like clinical documentation or claims management.

For the modern healthcare startup, technical excellence is no longer enough to win a contract. Success now depends on "architectural compatibility"—the ability for a new tool to slot cleanly into an organization's existing cloud environment without creating new security risks or data silos. As foundation models become the "operating system" of the hospital, vendors who try to bypass these established layers risk being sidelined by platforms that offer native, integrated AI capabilities.

Healthcare systems are currently trapped in "pilot purgatory," a frustrating cycle where dozens of disconnected AI tools are tested but never fully integrated into clinical workflows. Keith Figlioli (managing partner at LRVHealth) argues that the traditional software buying strategy—one tool for one department—is fundamentally broken in the age of generative AI. To achieve real return on investment, leaders must stop shopping for features and start designing a cohesive enterprise architecture.

This architectural shift relies on a three-tiered stack. At the bottom lies the core infrastructure of cloud providers and "systems of record" like Epic, which handle the heavy lifting of data and security. The middle layer consists of foundation model platforms that act as a control plane, managing how models interact with proprietary data. Finally, specialized startups sit at the top, solving specific problems like clinical documentation or claims management.

For the modern healthcare startup, technical excellence is no longer enough to win a contract. Success now depends on "architectural compatibility"—the ability for a new tool to slot cleanly into an organization's existing cloud environment without creating new security risks or data silos. As foundation models become the "operating system" of the hospital, vendors who try to bypass these established layers risk being sidelined by platforms that offer native, integrated AI capabilities.