How to handle serverful resources when using ephemeral environments

The author differentiates between "serverful resources", like RDS and OpenSearch clusters, and "things that should be a platform team's responsibility", like VPCs.

I'm currently running a shared RDS cluster, for instance, that's used in a platform capacity. Workloads are isolated from each other; there's a clear pipeline for my team and others to update it; and the eventual goal is to build out monitoring to make it evident which workloads are putting stress on the system, so that the owners can respond. We chose this architecture to save on infrastructure costs, at the expense of marginally higher operational costs, as we explicitly wanted to avoid a proliferation of tiny RDS clusters with each new service. The expectation is that the higher operational cost will ultimately be lower than what infrastructure costs would be.