Clinical documentation is one of the biggest bottlenecks in healthcare AI adoption: the data is messy, the stakes are high, and “great” model outputs often fail to translate into real clinical action. In this talk, I’ll share what we’ve learned building and deploying clinical documentation AI in live care settings – what works, what breaks, and why.

Using chronic care for older adults as a case study (primary care and post-acute/SNF workflows), I’ll walk through the end-to-end process of turning unstructured notes and medical records into usable outputs for care teams: evidence extraction, clinically grounded summarization, and workflow-ready actions (e.g., care gap identification, suspect gap review, documentation support, and care plan updates). I’ll cover the practical implementation realities that teams underestimate: data quality and fragmentation, ambiguous clinical language, varying documentation styles, oversight and accountability, and the human-in-the-loop design choices that determine whether clinicians trust the system.