Modern AI agents promise to revolutionize legacy modernization, yet most enterprises are dumping millions of lines of code into LLMs and watching them fail spectacularly. If you've felt the frustration of AI hallucinating through your 10+ year-old codebase or producing unusable suggestions for systems where business logic drifted from documentation years ago, you're experiencing what I've seen at multiple enterprises. But what if we could orchestrate specialized agents that actually understand both code structure and business domain reality?

In this talk, you'll learn how to move beyond single-agent approaches by implementing multi-agent architectures with distinct personas - Code Explorers that map system boundaries, Domain Experts that identify business concept drift, and Technical Product Managers that maintain context. You'll discover practical patterns for transforming event storming outputs into agent-readable UML/Mermaid formats and why traditional RAG approaches explode on legacy code. Most importantly, you'll understand how to implement "minimum viable context" indexing that gives agents exactly what they need without overwhelming token limits, pointing toward the next generation of agentic AI for modernization.