A look at the evolution of LLM tool-use, from supervised fine-tuning to Reinforcement Learning (RLVR) and agentic applications in large and specialized models.