LinkedIn engineers describe how they migrated 4+ million CPU cores across 2,300+ stateless services from a legacy platform to a Kubernetes-based compute platform using Temporal for workflow orchestration. The system was built around three non-negotiables: site availability, engineering productivity, and full automation. The architecture comprised three stages: application inventory (periodic fleet snapshots), pre-qualification automation (containerization and dry runs in staging), and migration automation (gradual region-by-region traffic ramp with health checks and rollback workflows). A supervisor hierarchy managed concurrent migrations at scale, with a supervisor manager spawning per-service supervisor workflows that coordinated multi-region ramps. Key design decisions included idempotent activities, stateless supervisors, GitHub PR-based consent for production services, durable timers for ramp pauses, and an emergency kill switch via custom worker interceptors. Temporal's durable execution, built-in scheduling, and UI observability were central to making the system safe and reliable at scale.