This presentation dives into the challenges of migrating enterprise Gateway configurations from NGINX to Envoy in an environment with tens of thousands of services across hundreds of clusters and regions. Migrating configurations is often seen as a simple translation task, but real-world scenarios reveal vastly different data models, incomplete or inconsistent information, and the need for common-sense interpretation. We introduce an agentic approach powered by Large Language Models (LLMs) to drive and refine migration tools iteratively, but supervised by a human-in-the-loop. The method leverages LLMs not just to transform configurations but to also fix and improve the migration tools through a structured inner loop (coding) and outer loop (migration execution) architecture. Attendees will learn how this innovative approach enables scalable, predictable, and efficient migrations with iterative human evaluation and agentic workflows.