How should Kubernetes autoscaling be implemented for production?

Choose metrics aligned to demand, define sensible minimums and maximums, tune stabilization windows, ensure Pods start quickly, and coordinate workload scaling with node capacity.