In reality what will happen is this. Within the next 5 years or so people will be abstracted to only working with the agent window or similar and never seeing code of anything. This will be how they interact with it and while they will succeed at many things the knowledge of precisely how data loads into something, for example, will be lost to the greybeards who had to do it prior to AI.
Low level tasks are completely destroyed by AI and it creates the problem we have today. You do not get Senior engineers without hiring Junior engineers.
It will create a point of finality. If you get to a point where you cannot prompt your way out of a massive defect what do you do? AI is only as smart as your ability to tell it what to do. Especially where critical defects and bugs do not necessarily result in a programmatic error but the inability of the application to do what it is supposed to within the parameters required.
As a better example. At my previous job we had archaic data architecture from the 90s running critical aspects of the business. Over time this had been layered with more than 80 transformation steps built into MSSQL to meet growing business needs. This still worked, mostly, but its processing time was being dragged out by essentially O notation and complexity. You cannot turn this system off, it runs the business, and its scale is too large to replicate to build a corrected parallel solution. You have to fix it in place.
On top of that, none of the code running this is in one singular coherent repo or anything either. It's all old databases that predated source control.
How do you think AI could even diagnose that as it can't see all of it?
