Use 100% automated fixes for known problems that don't require manual intervention. Good for known problems with known workarounds.
Kubernetes is fast-moving. Don't miss the chance to debug something just because you weren't online when it happened. Define what evidence should be gathered and let an automated runbook fetch it at the right time.
Define best practices for handling and fixing alerts. Encourage users to respond according to plan and not with ad-hoc fixes. Using automation, make the best fix the fastest one too.