Hospital for Special Surgery, the New York-based orthopedic academic medical center, now processes 1,100 insurance claims a month with AI agents and has cut appeals handling from 45 minutes to five, with the success rate climbing from 65% to 100% in the nine months since rollout. The agents, built in partnership with enterprise developer Ema Unlimited, have let HSS pull claims work back in-house after years of relying on a third-party contractor to handle the volume. It is one of the most concrete operational results yet published for agentic AI in a US hospital.
The stakes for the sector are large. The World Health Organization projects a global shortfall of 11 million health workers by 2030, and KPMG says 68% of providers have already deployed AI agents in some form. A separate KPMG figure puts 84% of providers as comfortable handing specific decisions to those agents, a sharp reversal from the cautious posture the industry held when electronic health records first rolled out in the early 2000 s.
Ashis Barad, chief digital and technology officer at HSS, frames agentic AI as a different category from telehealth or remote monitoring, both of which improved access but left the administrative burden largely intact. AI agents, he argues, can handle nuanced cases without defaulting to a human escalation, retrieve information from clinical sources, and iterate over time. That is the gap previous waves of health-care digitization never closed.
Building on the claims work, HSS is now pushing AI agents into patient-facing roles. A 24/7 scheduling and triage service, accessible by web, text, or phone, uses conversational AI to ask clarifying questions about a patient's condition and then books appointments based on the clinician's specialty, location, insurance coverage, and availability. The agent is trained on HSS's internal protocols, policies, and care pathways.
Safeguards sit on top. Sensitive, complex, or uncertain scenarios are escalated to human specialists, every agent decision is auditable, and human staff can intervene at any point. Decisions about which workflows to automate are filtered through an internal AI subcommittee that Barad co-chairs with a senior nursing executive, with patient-facing agents reviewed far more rigorously than back-office ones. The data foundation is where most providers stall. Barad notes that something as routine as time to start surgery has had a different definition at every hospital he has worked at, a fragmentation that blocks agents from assembling the tacit, cross-source context.