Generative and Agentic Artificial Intelligence for Medical Coding and Billing: A Human-in-the-Loop Architecture and Evaluation

Medical coding and billing workflows in modern healthcare systems have grown increasingly complex due to expanding clinical documentation, evolving coding standards, and heightened regulatory scrutiny. These factors contribute to persistent error rates, administrative inefficiencies, and financial risk, placing substantial cognitive and operational burdens on healthcare professionals. This study aims to design and evaluate a human-in-the-loop architecture that integrates generative and agentic artificial intelligence to support medical coding and billing while preserving expert oversight. The proposed framework combines automated code suggestion, contextual reasoning, and workflow orchestration with structured human validation at critical decision points. The study employs a mixed-methods evaluation approach, incorporating architectural analysis, workflow performance assessment, qualitative expert feedback, and quantitative measures of accuracy, efficiency, and error reduction. Results indicate measurable improvements in coding precision, turnaround time, and audit readiness when compared to conventional manual or fully automated pipelines. At the same time, the evaluation reveals residual risks related to model hallucination, edge-case misclassification, and workflow overreliance, underscoring the importance of continuous monitoring and human intervention. Overall, the findings demonstrate that human-AI collaboration offers a more reliable and accountable pathway than full automation for high-stakes healthcare administration tasks. The study concludes that strategically designed human-in-the-loop systems can enhance operational performance while maintaining compliance, transparency, and clinical trust in medical coding and billing environments.