UTMB Deploys 22 AI Agents Across Clinical Workflows
UT Medical Branch uses 22 AI agents for triage and imaging, aiming to extend staff capabilities rather than replace them.
Key points
- UTMB has deployed 22 AI agents to handle tasks like triage and imaging analysis.
- The system uses Carebricks, a platform also adopted by major health systems like Mayo Clinic.
- AI is positioned as a workforce extension for context management, not a decision-maker.
UT Medical Branch in Galveston, Texas, has deployed 22 AI agents across its clinical operations. These tools handle tasks like patient triage and lung nodule detection. The health system aims to extend its workforce rather than replace human staff.
Peter McCaffrey, the chief digital and AI officer, describes this as a platform approach. He argues that AI should fit into existing workflows to manage context efficiently. This strategy helps surface critical data that might otherwise be missed by busy clinicians.
AI extends clinical workflow capacity
The agents manage specific tasks like radiology follow-ups and pharmacy prior authorizations. They also support clinical documentation integrity and supply chain management. By handling these administrative and analytical steps, the AI frees up doctors for direct patient care.
One early use case involved screening chest CTs for coronary artery calcification. This biomarker indicates heart risk but is often overlooked unless specifically ordered. The AI systematically scans existing images to flag this data for physicians without extra effort.
Platform choice and governance structure
UTMB uses Carebricks, an agentic AI platform from Bunkerhill Health. Other major health systems, including Mayo Clinic and Cleveland Clinic, also use this technology. The choice reflects a move away from single-purpose tools toward integrated systems.
Successful deployment requires strong governance and alignment with safety values. McCaffrey emphasizes engaging service line owners early in the process. This ensures the AI supports equity and organizational goals rather than just adding new technology.
Trade-offs in automated decision support
The AI does not make final clinical decisions. It acts as a workforce extension for context management and synthesis. This reduces the risk of patient harm caused by missed or buried information.
However, reliance on AI requires trust in its underlying data. The system must be accurate and transparent to maintain physician confidence. Balancing automation with human oversight remains a key challenge for health systems.
