How to get Kubernetes the missing metrics for more efficient AI scheduling

AI is speed running virtualisation and cloud computing, but how do you isolate for security and still get enough information to make good decisions?