Search

The integration of generative artificial intelligence into clinical documentation workflows promises substantial efficiency gains yet introduces persistent misalignment between AI-generated drafts and individual clinician judgment. This conceptual systems research article advances a novel human-in-the-loop adaptation theory that explicitly models clinician preferences as dynamic, context-sensitive inputs rather than static constraints. Drawing on peer-reviewed evidence, the manuscript synthesizes how preference elicitation, real-time adaptation, and closed-loop governance can transform AI-assisted drafting from a supplementary tool into a co-evolutionary clinical intelligence infrastructure.Central to the contribution is the introduction of the clinician preference orchestration and adaptation framework (CPOAF), a layered architectural model featuring four interdependent strata and a star-topology feedback mechanism that propagates preference drift signals radially from peripheral clinician nodes to a central orchestration engine. Three interpretive mathematical constructs—decision confidence, monitoring burden, and drift sensitivity—are formalized to guide theoretical deployment without empirical benchmarking.The framework addresses governance constraints, data-modality specificity, and deployment-environment heterogeneity while preserving clinician autonomy. By foregrounding preference modeling as the core adaptive mechanism, CPOAF offers a scalable infrastructural blueprint for next-generation AI-assisted drafting systems that remain clinically grounded, ethically defensible, and institutionally sustainable. Implications extend to health-system informatics, regulatory science, and human-centered AI design.

Non-urgent patient portal messages consume a substantial fraction of clinician time and can shift attention away from higher-acuity care. As portal communication becomes a routine part of ambulatory medicine, inbox management increasingly functions as an additional clinical workload. Current inbox workflows often depend on manual review, interpretation, and free-text reply by clinicians. This creates delays, variability, and cognitive burden, especially for routine requests that could be answered through standardized guidance. This article proposes a large language model agent that classifies incoming message intent, identifies non-urgent queries suitable for templated responses, and drafts a structured reply. The draft is grounded in local clinical protocols, physician-approved templates, and patient-specific information from the electronic health record. The proposed agent includes a message intent classifier, a protocol-and-template retrieval module, a retrieval-augmented drafting model, a rule-based safety filter, and a human-verification interface. These components work together to keep generated text within a clinically approved and auditable workflow. By preparing protocol-adherent draft responses for clinician review, the agent could reduce routine documentation effort while preserving clinician judgment. Its value depends on safety boundaries, template quality, patient-data integration, and seamless fit within the existing portal workflow. The agent represents a pragmatic near-term use of large language models in clinical administration. It supports automation without removing human oversight from patient-facing communication.