Correct me if I'm wrong, but sounds like you are thinking of welding in a factory?

I know a lot of welding is repairing damaged equipment. Especially older or already Jerry-rigged devices. I'd imagine it would be extremely difficult and expensive to develop an AI and a machine that would be adaptable enough to do these sort of projects.

Would construction welders be any less 'non-standardized' than Plumbing or HVAC? I would think welders have to problem solve far more than them for their tasks..

Ah that explains it - a maintenance vs construction thing - notice how repair has become the expensive option compared to creating new? Manufacturing has a far more constrained process which allows for readily automatable braindead repetition because it gets results. If construction now isn't readily automatable surprisingly doesn't matter too much. Building new things can be reengineered to a "friendlier" option like say screws and nails instead of joins but maintenance needs to take what they have and their moving parts.

Building everything to a schematic is easier to automate but I believe that already isn't the bulk of labor demand.

Theoretically a smart system with diagnostics could be more readily automated but those would be more expensive and probably be new constructions and come with their own host of technical and logistical issues. Sure it would be nice to have clogs and leaks mapped and remedied but it would take a while to become universal even if they were rolled out tommorow and worked perfectly.