Something that's easy to miss: cost tooling and observability data can feed into the same loop, not two separate concerns

CloudCostTree reads your Terraform/CloudFormation/Pulumi and shows a cost tree from static config alone. That's useful before deploy, but static config only tells you what something was provisioned for.

Once it's actually running, the same tool can pull real CloudWatch metrics, CPU utilization, live Spot pricing, through your own read-only AWS credentials, and re-run the same analysis against real telemetry instead of declared instance types. Same report, same command, just backed by observed data instead of assumptions.

Point being, a resource can look perfectly sized on paper and still be sitting at 4% CPU in production. You only catch that gap by closing the loop between what you declared and what you're actually observing.