As digital pathology matures beyond image capture toward workflow optimization, pathologists have a timely opportunity to shape how AI is integrated into routine practice. A novel opportunity is now emerging: modern whole-slide scanners ship with meaningful on-device compute, making it possible to run AI inline during acquisition rather than after the fact in the cloud or on a separate server. We describe an edge-AI architecture built on Virchow foundation-model embeddings deployed on scanners with this local compute. Lightweight, task-specific attention heads are trained and validated entirely on-scanner; no patient data leaves the institution at any point. Because the foundation model carries most of the representational burden, each new use case can be supported with a small, locally curated slide set and a modest training run. Unsupervised clustering over the embedding space helps surface morphologic patterns and guide hard-negative selection. This architecture underpins FROST, our frozen-section triage pipeline. Basal cell carcinoma detection on frozen sections has been validated inline on the scanner, showing that a locally trained pipeline can deliver triage signals in real time without disrupting throughput or moving data off-site. Building on that foundation, we are now working to validate and deploy dysplasia detection on head-and-neck frozen sections for intraoperative triage, and are testing the same approach for colon and breast cancer triage on routine biopsies. Outputs drive prioritization and ancillary-study pre-ordering rather than autonomous diagnostic calls, keeping pathologists in control of interpretation. A guiding principle is to focus edge AI on high-value, low-risk applications: tasks where AI augments rather than replaces judgment, and where even gains in turnaround translate directly into patient benefit. Triage and prioritization sit squarely in that space, offering a practical, locally governed path to AI-augmented pathology.

Learning Objectives:

1. Explain how on-scanner edge compute enables inline AI deployment while preserving institutional data governance
2. Describe FROST as an edge-AI pipeline validated for basal cell carcinoma and extending to dysplasia and cancer triage.
3. Articulate why high-value, low-risk applications such as triage are the right starting point for edge AI in pathology.