The Augmented Pathologist: Agentic Workflows in Practice

As digital pathology continues to expand across institutions, the conversation is shifting from infrastructure to impact. Whole slide imaging is no longer the finish line-it is the foundation. The next phase of transformation involves building workflows that are smarter, faster, and more resilient through agentic systems: task-oriented AI agents that can perceive, plan, and act, all while keeping the pathologist in control. Our experience implementing these workflows spans clinical and research environments. We have developed image analysis pipelines and integrated foundational vision models to autonomously triage cases, quantify biomarkers, and generate LIS-compatible synoptic reports. These components serve as essential building blocks for future agentic workflows, enabling scalable and adaptable automation across diagnostic contexts.We are also advancing agentic capabilities through multimodal data integration. By incorporating eye-tracking data, clinical metadata, and workflow signals, these systems can interpret not only static image content but also user behavior and diagnostic context. In pilot implementations, we have introduced stream-of-consciousness dictation, allowing pathologists to narrate findings as the agent generates draft reports, flags ambiguities, and suggests stains or molecular tests. We are exploring multimodal foundation models capable of processing audio and video inputs to provide more context-aware support, aligning with real-time diagnostic reasoning while minimizing disruption to workflow.Education is a core component of this agentic ecosystem. These tools offer new approaches to teaching pathology, particularly in introductory histology, where agents can guide learners through key features, reinforce pattern recognition, and connect morphologic findings to diagnostic concepts. Agentic systems enable focused, case-based teaching by adapting content based on user interaction and performance. Eye-tracking and behavior data help identify areas of uncertainty, allowing the system to deliver targeted feedback and reinforcement. These same platforms can also support the development of assessment questions grounded in real diagnostic tasks, offering insight into trainee progression and curriculum refinement. We will outline deployment strategies, discuss governance and safety considerations, and share concrete examples of how agentic workflows are reshaping the daily practice of pathology-not by replacing the pathologist, but by supporting them with intelligent systems that enhance accuracy, efficiency, and education.

Learning Objectives: 

1. Describe the core components and capabilities of agentic workflows in digital pathology.
2. Explain how multimodal data integration including eye tracking, voice input, and foundational models can enhance diagnostic support.
3. Identify practical strategies for implementing and governing agentic systems in clinical and educational pathology settings using existing infrastructure.