Key Takeaways

  • Epic’s Connectors allow its AI assistants to retrieve information and initiate actions in external applications.
  • The capability could reduce workflow fragmentation across clinical, patient-service and administrative systems.
  • Healthcare organizations will still need to address access controls, data quality, legacy integration and oversight.

Epic has launched Connectors, a capability that enables third-party applications to exchange information with its artificial intelligence assistants and respond when those assistants request an action. The launch expands the company's AI strategy beyond answering questions inside its own environment. Its agents can now participate in workflows involving external systems used across hospitals, clinics and related operations.

The capability applies to Emmie, the patient-facing assistant; Art, its assistant for clinicians; and Penny, which supports revenue-cycle and operational work. It also supports custom AI created through Agent Factory. In practical terms, a connector can give an assistant access to an approved external function, subject to the participating applications’ configuration and the healthcare organization’s governance policies.

That distinction matters. An assistant that only summarizes information has a narrower operational role than one that can initiate a transaction. Connectors could support tasks such as placing a patient meal order, providing hospital wayfinding information, coordinating an ambulance arrival, checking medical-equipment availability or helping manage pharmacy logistics. VPLRx and hospital dietary-management applications are among the named ecosystem examples.

Healthcare organizations already operate large collections of specialized applications, many of which handle a small but important part of the patient or administrative journey. Staff members often move between those systems manually. Allowing an AI assistant to carry context from one application to another and request the next step can reduce handoffs and repetitive data entry, but it also introduces a larger integration and oversight burden.

That burden is hardly theoretical. MIT Technology Review Insights research reported by HealthManagement.org found that 98% of surveyed healthcare leaders considered interoperability a continuing challenge, although 59% described it as difficult but manageable. While 96% said their organizations were ready and resourced to implement digital-health solutions, only 26% characterized themselves as “very ready.” The gap suggests that executive interest may be running ahead of technical and operational maturity.

Public-sector healthcare faces a similar mismatch. A Market Connections/Google survey reported by GovExec found that 65% of federal healthcare respondents expected AI and machine learning to have the greatest effect on interoperability over the following five years, yet only 17% reported full AI and machine-learning integration. Meanwhile, 66% identified legacy-system integration as a leading technical challenge, and the same percentage reported data silos.

Standards can help, though they do not resolve every workflow issue. HL7 FHIR APIs provide a common approach for structured health-data exchange, while TEFCA and Common Agreement 2.0 support broader nationwide information sharing. The Office of the National Coordinator for Health Information Technology also oversees a certification ecosystem that supports standardized APIs and USCDI data elements such as medications, laboratory results, allergies and clinical notes.

Still, connecting an AI assistant to an application involves more than making data readable. Health systems need to decide which actions an assistant may initiate, what approvals are required, how identity and authorization are maintained, and where activity is logged. A meal order carries a different risk profile from a pharmacy transaction or a clinical workflow. Connector deployments will likely need permissions tailored to each use case rather than broad access.

There is a competitive angle, too. By giving outside developers a route into Emmie, Art, Penny and Agent Factory, Epic is positioning its assistants as an orchestration layer across the wider healthcare application ecosystem. Third-party vendors gain another channel through which users can reach their functions. The company, in turn, can make its AI assistants more useful without building every specialized application itself.

For healthcare technology leaders, the immediate question is less whether connectors are interesting and more whether their integration foundations are ready. Organizations with mapped workflows, reliable APIs, clear ownership and mature access controls may be able to move faster. Others may find that AI exposes familiar interoperability weaknesses in a new form. The launch broadens what healthcare agents can do, but the value will depend on disciplined implementation, not simply switching the connections on.