Over 10 years working with Java enterprise systems, I've been through some major cloud migrations and watched two fail from the inside. This talk shares the architectural decisions that never make it into the case studies: how to strangle a legacy Spring monolith without breaking production, why database decomposition is more dangerous than service decomposition, the exact point where microservices create more problems than they solve, how to handle the shared authentication state that your monolith has accumulated over a decade, and what "done" actually looks like when you have 200 services in Kubernetes.