Agentic AI workflows in anatomic pathology: Opportunities and challenges

Abstract

Artificial intelligence (AI) in anatomic pathology is dominated by image-based diagnostic applications, but much of the specialty's cognitive work is textual: reviewing clinical history, drafting reports, coordinating testing, and communicating results. Large language models can be useful for many of these tasks, and agentic systems extend their capabilities by retrieving external information, invoking software tools, following task-specific instructions, tracking workflow states, and carrying out complex multi-step tasks under human oversight. This conceptual review outlines key components of agentic AI systems including retrieval augmented generation, tool use, skills, and the so-called “agentic loop”—and applies them to anatomic pathology workflows. Three illustrative design sketches use reflex testing support, workload balancing and case routing, and ancillary test decision support to explain how agentic systems can combine lab information system data, trustworthy knowledge sources, local policy, and explicit human approval gates. We examine considerations for clinical deployment, including interoperability and integration challenges, lifecycle governance and validation, security and safety concerns, and cost and sustainability. Agentic AI offers a plausible and constructive near-term vision for AI in pathology: better-supported diagnosis that allows pathologists to spend more of their time on the judgment, accountability, and genuine care that remain distinctly human.