This is not that much different in programming, it just takes longer to find out.
Using AI is like using adaptive cruise control: you're still in charge of where the vehicle will ultimately end up and if you decide to go take a nap you will wake up in the hospital or end up in the morgue.
It's true though that the 'time between interventions' is lengthening steadily.
The major difference is that people routinely ship shoddy, broken products in the world of software with few consequences, beyond needing to patch it later. Try the same thing with a court filing and you’re out on your ass.
>This is not that much different in programming, it just takes longer to find out.
The primary difference is that it's much easier for the AI to figure out itself if it's wrong on the programming side. Knowledge workers don't have automated testing for example.
The end result is the knowledge worker needs to be more heavily involved in the process and be more mindful of mistakes which makes it a worse experience for the user and generally less productive than it is for programmers.