Generative AI has evolved from basic chatbots into agentic systems that can do more than hold a conversation. This talk explores what it takes to build a production-ready AI agent that answers questions, understands context, uses tools, and takes action to solve real problems. Even the smartest LLM is limited without access to up-to-date knowledge or the ability to interact with the world. We'll look at how adding memory, external data, and the ability to take action can turn a passive chatbot into a proactive assistant.

We'll cover the building blocks of modern AI-powered software. Techniques like Retrieval-Augmented Generation (RAG) and targeted tool use help ground AI in real-world context to improve precision and reliability. We'll also look at open standards like the Model Context Protocol (MCP), which standardizes access to external tools and data. Along the way, we'll discuss how to balance speed, accuracy, and cost when designing systems for real-world use.

We'll also cover prompt and context engineering, which help guide model behavior and reduce hallucinations by providing the right information at the right time. You'll leave with tools and ideas for building AI-powered software that goes beyond a wrapper around a mainstream LLM.