For anything in the top ~1% of the human social ladder - you can count on the politicians, plutocrats, etc. who make the rules in human society being extremely reluctant to allow AI's to start replacing their own kind.
Your "can't ever" is too ambiguous. If "ever" allows thousands of years, then it's purely a philosophical/linguistic question. If you're mostly interested in credible threats to your own career, then it's more a technology/society question.
Punishment with animals, as other species, doesn't work as well because it lacks information about your intent, and presumes the animal has understood the intent it is being punished for. Negative reinforcement works in only very limited and immediate circumstances where it's designed to set a memory - but you don't get much influence in the memory you are setting with it, you just get pain and an association of their own experiencing. In the case of an electric fence, you don't care, you just want them to respect fences (though I work with a stallion who balks at fences now because he was in a paddock with electric fences to keep him from knocking them down), but if you wanted to put electrodes on their body to send them in directions, you would need a neuro-feedback reward mechanism. In the case of training dogs and horses, that feedback mechanism is the emotional approval and praise, which builds rapport and a way of relating.
In all but actively malicious situations, even most people don't understand what they are being punished for, let alone animals, and it's a neutralizing mechanism. Using ML/AI for stimulus/response training seems like a risk for a lot of cruelty to train the model, and even where it uses rewards, the deprivation and dependency becomes a kind of cruelty as well. I don't know there is an ethical way to do this.
* Any tiny leak of liquid or gas is a huge problem and liability, thousands of units of resources (water or whatever) can be wasted and also cause property damage or kill people (gas leaks)
* Figuring out and sourcing the right fittings and BoM can be challenging
* It's fundamentally a messy business, mixing primer and glue, digging trenches, hard to reach spaces, encountering other unknown pipes or obstacles and needing to figure out head scratchers like "well how important is this unmarked X in the way of where I need to route"
* Lots of jobs involve retrofitting the future thing into existing (old, or even very old) structures where everything is variable from one case to the next
In many ways, programming is a lot easier since the complexity is in your head and not a physical problem.
In general I love DIY and do it for a lot of things, but when it comes to plumbing the lessons I've learned are 1.) when possible, leave it alone and not mess with it, and 2.) Otherwise bite the bullet and hire a professional.
Some of those things sound like something a machine will become much better at than a human ever could.