Building a single AI agent is easy, but engineering a platform that can support thousands of them is a different story. Have you ever tried to figure out which department is spending the most on tokens, or how to ensure every agent that ends up in production follows your safety guidelines? If you are struggling with shadow AI, you aren't alone. In this session, inspired on the Citadel Foundry Platform, we will show you how to build a centrally governed AI platform on Azure. An AI engineer and an Azure platform engineer team up to walk through a reference architecture that handles token rate limiting, semantic caching, and security through APIM and Microsoft Foundry. We will look at how to manage agent identity with Entra ID and how to deploy this entire setup as a hub-and-spoke landing zone using Infrastructure as Code. On the practical side, we cover implementing an AI Gateway on APIM, setting up agent identity with Entra - everything deployed as code. Join us to see how you can move from scattered AI projects to a professional, manageable platform that your organization can actually trust.