The issues here are there is an assumption AI will improve so far whereby it has the capabilities to write flawless code for pieces of software that would be many separate files and many thousands of lines of code. There is very little about LLMs that suggest there is any focus on accuracy to this level or that it would be possible within any foreseeable future as we are currently are nowhere near this at all.

Secondly, we use software and AI as a human race that others have created as many are not quite aware of what they need and don't have the capabilities of designing software to solve all of the issues they have. That won't suddenly be a skillset many pick up.

> Why do I need a legion of Salesforce consultants to make their shit work with my shit when I can just have the AI write all the reports I'll ever need?

The article has some interesting thoughts but I think this one misses the mark.

Reports aren't the hard thing, it's the overall system that is modeling complex business state and state transitions. This business state is frequently at an abstract level above the individual enterprise systems because it's spread out between multiple systems.