This talk explores the transformative capabilities of Large Language Models (LLMs) like Gemini in powering intelligent agents, beginning with an overview of their advanced reasoning abilities shaped by post-training techniques. We delve into how context, memory, and augmentation strategies such as Retrieval Augmented Generation (RAG) and function calling expand LLM functionalities, enabling them to become effective agents that can reason and act. The session will further discuss the evolution from single to multi-agent systems for tackling complex problems, differentiate between predefined workflows and dynamic agent behaviors, and introduce the Agent Development Kit (ADK) as a practical framework for designing and implementing these sophisticated AI systems, culminating in a demonstration of ADK in action.