The AI will be pushed everywhere, but as it's currently not succeeding at driving across the USA without a human around for emergency intervention* despite a trillion or so miles of real world experience, I suspect the initial success of AI in plumbing looks more like "humans being meat robot e.g. by wearing AR headset while physically at customer site" and less like "AI controlling metal robot that is physically at customer site".

That said, the economics of AI… eh. I can see it replacing the $3T/year cost of hiring software developers, but this doesn't mean people will spend $3T/year on AI; I mean, Wikipedia killed demand for the Encyclopaedia Britannica, but nobody values Wikipedia at the 16 trillion or so USD you'd get by multiplying the cost of the EB with how many people are online.

* even the CyberCab, which was announced as "no steering wheel or pedals", is geofenced and if it's like their other vehicles has human observers on standby for remote control.