“I was using our same new deployment process on EKS, and I would set, basically, a base size and then have a node group that had a ton of space to scale up and down with using the cluster autoscaler built into Kubernetes. So that when the workload came online, it would say, give me more instances. But it was in large chunks of instances that it would do that versus Cast AI. When we need a little bit more space, Cast AI gets us a little bit more space. When we don’t need that space, that instance goes away. I couldn’t replicate that without building my own version of Cast AI.”