In this video, I build a multimodal AI agent that can actually “understand” and analyze media—YouTube URLs, normal uploaded videos, and images—and then chat with you about what it sees. The core idea is simple: instead of treating video like a black box, I show you how an agent can extract the key information from a YouTube link, generate summaries automatically, and answer questions in a way that feels like you’re talking to the content itself. I walk through the full demo first so you can see the end-to-end behavior: you give the agent a YouTube video (or your own video), it processes the content, and you can ask targeted questions to get accurate answers. Then I break down the build step-by-step with an agentic/system-design mindset—what the agent needs to do, how it decides the next action, and how you can extend the same pattern to other media analysis projects. If you’re new to agents, I also point you to my prerequisites on AI agents and agentic design patterns so you don’t get stuck on foundations before shipping the project.