I think runbooks only make sense when the cost of automation is high, the task is infrequent, and the opportunity for / blast radius of an operator error is limited.

Deployments have never met those criteria for me, but there’s always something.