Writing a Kubernetes monitoring script is squarely in AI's wheelhouse.

The task is structurally identical every time: query an API, parse output, compare against fixed thresholds, fire alerts. There is no judgment variation between runs.

Success criteria are explicit — restart count > 3 per hour, memory > 80% of request limit, Slack alerts, retry logic. An agent can verify its own output against these criteria without human interpretation.

The agent has full knowledge of kubectl, Bash, and Slack webhook patterns, but cannot access the live cluster, actual kubeconfig, or Slack webhook URL. The human must supply environment-specific values and validate the script in context.

A buggy monitoring script fails silently or sends false alerts — annoying but not destructive. It does not mutate cluster state, so errors are low-stakes and easily corrected.

The logic is purely mechanical: arithmetic comparisons and conditional branching. No taste, ethics, or relationship context is needed to write or evaluate this script.