In oncology, clinical decision-making (CDM) requires integration of multimodal data and multidisciplinary expertise. Advances Large Language Models (LLMs) offer new opportunities, yet generalist multimodal models remain constrained by computational demands, retraining needs, and narrow task focus. LLM-based AI agents provide a more efficient alternative, enabling multi-task orchestration, real-time tool access, and integration of heterogeneous data. Given the limited evidence in CDM, we propose a specialized agent for advanced non-small cell lung cancer (NSCLC) to support CDM in first-line immunotherapy (1L IO).