I think that automation is fine, but when you start talking about life-critical systems, the bar on failure modes and reliability becomes a lot higher. Like, if you have a sensor for excessive chlorine levels, the default probably should be not to keep pumping more of the chemical in, but to have some “hard” upper level where no more chlorine goes in (and if the two systems aren’t directly-integrated, to at least let the maintenance safety system power off the chlorine system), and for a secondary alarm to be triggered (like, start playing a loud audible alarm) if maintenance doesn’t respond to the email notification and indicate that they’ve dismissed the alert within a given amount of time. That’s disruptive and undesirable compared to just emailing maintenance, but as long as there aren’t any failures, you shouldn’t ever see it. But if you do have it come up, you don’t want the failure mode to be chemical-keep-being-dumped-into-the-system-and-people-being-unaware-of-it.

You want fail-safe behavior.

Automation can increase safety by decreasing time until an issue is addressed, for example. Automation isn’t intrinsically incompatible with safety. Aircraft computers today use TCAS to automatically instruct pilots what to do if they are about to collide (and in fact, today the automated system is explicitly given higher priority than human air traffic controllers — if the two give pilots conflicting directives, the human is the one to be disregarded). But you gotta make your automated procedures have high reliability and handle failure the way you want when the consequences of a failure are high (and that goes for non-automated procedures too).