Vanadium 50 said:
To lowest order, approximately 99% of the time and effort is spent figuring what you want to do, and 1% coding it up. AI can maybe speed up that second part.
There's a difference between personal software development and large scale (application) software development. In the latter case, a project team consists of many different specialists (from the client/users who specify the system, through business analysts, architects, designers, software devolopers, testers and implementation and support staff).
The figure that was bandied about when I was working that it took ten times the effort per function point in the latter case.
I've seen projects with hundreds of development staff working over several years.
The code generation engines that were available in the 1990's were abandoned and superseded for reasons that I never understood. It's years since I thought or wrote about this (I've been retired for over ten years), but I was suspicious that the techies in the IT industry had pulled the wool over everyone's eyes and kept IT as labour-intensive as possible.
It wasn't as simple as that, of course, but even before AI, I believed that the world would have been better served if more effort had gone into the automation of IT development.
AI could decimate the labour required for software development and support generally. In precisely the way that IT itself obliterated the need for many clerical jobs. Whether those IT people will move on to other jobs is a different question. But, IMO, there is no reason that the bulk of what goes on to develop software cannot be automated once AI is in the picture. Starting with requirements gathering and management!
PS if we are thinking radically, then many of the end users would be replaced as well. The requirements would be a communication between a business-specialist AI and a software-analyst AI and thence to the software development AI. That may not happen, but I wouldn't bet against it.
Note that for small-scale projects, where a small group of people can do everything from requirements specification to development, there is perhaps no advantage of AI.
I'd be cautious of going into software development as a graduate in 2024.