Everyone talks about modernizing legacy systems. Few people actually open a COBOL codebase and do it. In this session, I will. Using real mainframe applications, I'll walk through how GitHub Copilot's agentic capabilities can help you understand, plan, and extend COBOL code you've never seen before. We'll start where every modernization project actually starts: figuring out what the code does. Using structured prompt workflows, we'll reverse engineer modules, discover data structures, trace transaction flows, and build a knowledge base that turns tribal knowledge into something an AI agent (and your team) can actually work with. Then we'll go further: planning and implementing a new feature, end to end, with Copilot doing the heavy lifting while we stay in the driver's seat. This isn't a "replace your mainframe overnight" talk. It's a grounded, hands-on look at what's possible right now when you combine spec-driven development with agentic AI. You'll leave with a repeatable approach to legacy discovery and modernization, and a realistic sense of where AI helps, where it struggles, and where you still need to know your stuff.