Autonomous AI research agents in pancreatic ductal adenocarcinoma immunology: from multimodal evidence to experimental validation

Pancreatic ductal adenocarcinoma (PDAC) remains one of the most challenging malignancies due to profound biological heterogeneity, complex stromal organization, and limited responsiveness to current immunotherapies. Recent advances in single-cell sequencing, spatial omics, digital pathology, and multimodal profiling have revealed that PDAC immune states are shaped not only by cellular composition but also by spatial organization, treatment context, and patient-specific evolutionary trajectories. However, the increasing complexity and fragmentation of biomedical evidence create substantial challenges for conventional artificial intelligence (AI) approaches that are typically optimized for predefined analytical tasks. Autonomous AI research agents represent an emerging paradigm that extends beyond predictive modeling by integrating scientific reasoning, retrieval of heterogeneous biomedical knowledge, computational tool orchestration, workflow adaptation, and evidence traceability. In this Review, we examine how agentic AI systems may contribute to PDAC immunology research, from characterization of immune ecosystems and spatial cellular interactions to biomarker discovery, neoantigen prioritization, and experimental protocol design. We distinguish AI agents from conventional task-specific models and discuss essential capabilities required for reliable biomedical research, including planning, tool use, verification, memory, provenance tracking, and human oversight. Despite promising technical demonstrations in biomedical domains, direct evidence supporting fully autonomous AI-driven discovery and validation in PDAC remains limited. Therefore, we propose a disease-oriented evaluation framework emphasizing benchmark task design, appropriate human and computational comparators, reproducibility assessment, and experimental validation. Future PDAC-specific AI research agents should prioritize evidence-aware reasoning, multimodal integration, calibrated uncertainty, and transparent computational–experimental workflows. Such systems may ultimately accelerate mechanistic discovery and precision immunology, while maintaining rigorous standards for biological validity, reproducibility, and responsible scientific practice.