Bunkerhill lands $55M to put 20 AI agents inside a working hospital system

Bunkerhill Health just closed a $55 million Series B to scale Carebricks, an agentic AI platform that already has 20 custom-built agents running inside a major Texas health system. The round included Sequoia Capital, Felicis, and new backer Khosla Ventures. Vinod Khosla himself cut through the usual healthcare AI noise, saying the bottleneck was never the technology but getting a health system to actually run it. That’s the bet here.

The pitch isn’t about shinier algorithms. It’s about plumbing. Most machine learning pilots die in the sandbox, never touching a live patient chart. Bunkerhill’s platform lets hospitals like Cleveland Clinic, Intermountain Health, and UTMB build their own agents that operate against real clinical data. Some scan cardiology imaging for early heart disease. Others handle the soul-crushing administrative work—prior authorizations, registry updates—that burns out staff. The company frames the opportunity around a brutal math problem: $5.3 trillion in US healthcare spending and a workforce that can’t keep up.

UTMB offers the most concrete look at what happens when the pilot label comes off. Dr. Peter McCaffrey, the system’s Chief AI Officer, reports that a coronary calcium detection agent flagged a patient at imminent risk of a heart attack in its first month live. Cardiology confirmed it and performed a triple bypass. The care team credits the early catch with saving the patient’s life. It’s a powerful anecdote, not a peer-reviewed trial, and Bunkerhill hasn’t published data on false positive rates or broader population performance. Other UTMB-reported wins include cutting specialist wait times by more than 50% with a nephrology triage agent and an 80% faster response on urgent lung nodule follow-ups.

Those are operational numbers from one institution, reflecting its specific data and staffing setup. Another hospital might not see the same curve. The new funding is slated to expand Carebricks into more clinical and operational use cases while building out governance and monitoring tools. The underlying tension is clear: letting a nephrology department build its own triage agent also means that department owns the consequences of how that agent is tuned. Bunkerhill makes the scaffolding. The health systems still have to do the hard work of not screwing it up.