The Landscape of Computational Pathology Agents From Static Analysis to Autonomous Diagnostic Workflows

The evolution of Computational Pathology (CPath) is undergoing a fundamental paradigm shift from traditional static pattern recognition to autonomous diagnostic systems capable of clinical reasoning. While previous foundation models successfully addressed scalability through self-supervised learning on whole-slide images (WSIs), they remained passive tools lacking the agency to interrogate slides or navigate complex clinical workflows. To bridge this cognitive gap, the field is pivoting toward CPath agents that leverage Large Language Models (LLMs) as central cognitive cores to perceive, think, and act in an interactive problem-solving process. These agentic systems are defined by four pillars including multi-scale perception of gigapixel landscapes, strategic cognition through reasoning and planning, autonomous action through tool utilization, and the maintenance of long-term memory for longitudinal case tracking. Current research explores diverse architectural paradigms such as single-model end-to-end agents, multi-agent collaborative systems, and human-in-the-loop frameworks that align with expert intuition. Despite significant progress in developing specialized generative benchmarks and evaluation metrics, the transition to clinical deployment faces critical challenges regarding reasoning-centric data scarcity, computational efficiency in gigapixel processing, and the imperative for verifiable clinical safety. Ultimately, the future of the field lies in the development of trustworthy, human-centric collaborative agents that function as true diagnostic partners to pathologists.