This hands-on session will guide participants through the process of building LLM-driven multi-agent systems, focusing on how large language models can collaborate to solve complex tasks. We will begin by exploring simple, hard-coded agent interactions and discuss practical strategies-e.g chain of thought and reflection -for improving LLM performance in coordinated settings. The talk will then introduce AG2 (formerly Autogen), a powerful open-source framework for constructing fully autonomous, multi-agent workflows. I will demonstrate how agents can be extended with function calling to interact with external tools and environments, enabling them to reason, plan, and act beyond text generation. Throughout the session, I’ll present real-world examples that showcase the potential of multi-agent systems to automate various tasks, including coding, multi-modal analysis, scientific discovery, and dynamic data retrieval in complex environments.