The deploy should be boring

A deploy becomes an event when any part of it is uncertain. Not slow — uncertain. Teams will happily wait eight minutes for something that always works, and will schedule a Thursday-afternoon window for something that works four times out of five.

Most of removing the uncertainty is unglamorous. Every build produces an image that stays addressable, so rolling back moves traffic to something that already exists instead of sending you back through CI to recreate it. Rollouts are gated on health checks rather than on a timer, so "it deployed" and "it is serving" are the same claim. Build output stays attached to the deployment that produced it, so a failure three weeks ago still has its reason next to it.

None of that is clever. It is the accumulated result of deciding, once, that the interesting part of your week should be the code and not the pipeline.