It is not like we are anywhere close to the point where someone can say "I want a complete solution for X". Hell, I've never worked on a project where someone was actually able to fully articulate what they wanted/needed. I doubt anyone has. The inability to say in plain talk what it is that is needed is at the very core of the critique of waterfall design.
What AI is at the moment is essentially IDE code completion with code block generation. AI is not going to help your CFO create that boutique accounting system they want. You have to know how to program and design to even know what questions to ask of your AI tool. That is not going away anytime soon.
If your solution integrates with boutique APIs and libraries then at the moment coding AIs are a hindrance. They don't tell you when they don't know. They will confidently make some bullshit up. This also requires coding experience to know when AI is bullshitting you. I can tell you from first hand experience that AI code completion and GIS system integration is a waste of time.
It’s like worrying about nuclear war - low probability of a negative outcome completely out of your control.
Those were gentle transitions. Probably the biggest was from mainframe to PC. I was in a team of designers in the early 90s and found I was the only one who could think in terms of user interaction instead of a basically linear flow. I had worked on getting to grips with it.
My career is over 30 years long so far, and I've never felt this. AI isn't making me feel it now.
What is true is the exact landscape is in flux -- but that's been true for longer than I've been in the business. All it means is that it remains necessary to do what has always been necessary to do: engage in constant learning and skill expansion.
Overtime the abstraction level of programming has changed, earlier you would write a sort function in assembly, now you just call the standard library. This was bound to happen due to the idea of software/code re-usability. AI for programming, if it kept improving, will be just another abstraction layer in the same way but that doesn't mean anybody who has no idea whatsoever about fundamentals of software engineer and programming can just use AI to create (complex enough) software.