I push to the right branch and expect it to just go live.
I don't think about it anymore. I've seen the pipeline work so many times that I have a kind of blind trust in it, and that's probably the best thing you can say about a pipeline.
What shook that trust a couple of times wasn't the pipeline. It was the pace.
I can't type fast enough to do four or five deploys in a few minutes. The AI can. Then they queue up for hours, and the update ships very late, or the runner chokes and it never ships at all. That's more the AI's fault than the pipeline's, but the user can't tell the difference.
So the rule is: commit often, never push until I've said yes, and bundle a feature or a piece of work into one push when possible. It works well. Sometimes the AI itself suggests that something has to go out before something else can work, and then we make an exception.
Infrastructure is the most important part of the whole stack, and at best you don't notice it. I have great respect for the people who make a living from nobody seeing their fantastic work.