This may be a little trial and error, but we should set some level of sane and logical resource requests and limits on our workloads in kubernetes rather than having nothing set for most things. This is related to PTD owned stuff like mimir, loki, system level stuff running from helm charts, team-operator, etc.
Not counting workbench session pods, those do have values set and there's a separate ticket for optimizing them.