Abstract. Accurate and timely identification of malignant morphology in histopathology slides remains a major challenge and is still largely performed by expert pathologists scanning whole-slide images for nuclear atypia, crowding, and architectural distortion. Conventional deep learning systems can detect cancer-associated patterns but are typically trained in a static “train-once/predict-once” fashion, behave as black boxes, and offer limited support when confidence is low or domain shift occurs. To address these gaps, we present Look, Mark, Explain, a minimal agentic AI framework that combines classical computer vision with large language model (LLM) reasoning to support cancer slide review while keeping the pathologist in the loop. High-resolution H&E images are processed with a contrast-enhanced, marker-controlled watershed pipeline to segment nuclei and extract 27 interpretable features per cell capturing size, shape, chromatin texture, nucleoli, and spatial context. For user-selected nuclei, a multimodal LLM receives a cell close-up, a contextual tissue view, and the feature vector, and returns a cancer-likeness assessment, confidence, and a pathology-style explanation that explicitly links its judgment to the provided measurements and visible cues, without issuing a formal diagnosis. We demonstrate the system on a breast cancer sample with 971 segmented nuclei, showing that population statistics, size-stratified comparisons, clustering metrics, and LLM narratives align with classical cytologic criteria and highlight enlarged, hyperchromatic, irregular cells in crowded, disorganized regions. The prototype uses standard digitized slides without stain normalization and is embedded in an interactive interface for slide exploration and on-demand AI consultation. While not intended for clinical use, it illustrates how a simple, reasoning-first, pathologist-in-the-loop agent can “look, measure, and explain” in a way that complements existing workflows and provides a testbed for future human–AI studies in digital pathology.