RAG has transformed AI systems by combining retrieval and generation, but traditional workflows often struggle with the dynamic demands of real-world applications, such as multi-step queries or integrating external APIs. Agentic behavior enhances RAG by enabling LLMs to make decisions, call tools, and adapt workflows dynamically. In this talk, we’ll define agentic behavior, explore its core features such as routing, tool integration, and reasoning, and demo its practical implementation in Python using Haystack.