> AI has failed to automate driving. (Despite enormous efforts)

Sure, if you consider this an all-or-nothing boolean, the OP is right.

However, AI has automated many driving tasks and outperforms the average human: take parallel parking, for example. Not an incredibly difficult task, but it's one where our modest AI absolutely outperforms the average human.

Similarly, the precursor to AI, autocomplete, can at times outperform the average human programmer. AI-enabled autocomplete (e.g., GitHub Copilot, etc.) almost consistently can outperform the average human programmer.

Defining the goal as "AI completely replacing programmers" is not a useful goal.

Defining the goal as "AI making the below-average programmer useful if they learn to use AI-enabled tools, and making the below-average programmer obsolete if they refuse to" is very pragmatic, and AI is practically there.

This is an interesting analogy. But I would like to propose a slightly more relevant comparison than “driving” versus “programming.”

In the 70’s and 80’s, electrical and electronics engineering was in high demand. Why? Because just about every product development area required digital electronics, analog electronics, or some hybrid of those. This was before FPGA, capable microcontrollers, SoC, touch screens, the World Wide Web, etc. EE’s were commanding crazy salaries for specialized skills.

PC’s weren’t commonly used for design. It was done by hand, on paper, from design books, data books, magazines and application notes.

In the late 80’s, 90’s and early 00’s, all of that started to change. Computers got more sophisticated, operating systems got more sophisticated, and design tools became available. “Great!” said EE’s, “now we can still make big salaries and we don’t have to do everything by hand!”

And then it happened: computers (microcontrollers) and FPGA became sophisticated/fast/cheap enough to replace most digital circuits; and, large scale integration allowed for SoC’s. All you needed to do now was to “tell” the computer or FPGA what you wanted it to do. Or, you just stuck a little glue circuitry around a SoC.

Now, we call this “telling” a computer what to do “programming,” but originally it was just a shortcut for hard design.

Programming became so prevalent that an entire career discipline grew up with it: computer science and the more advanced form, software engineering.

Of course, EE’s continued to program circuits, microcontrollers, larger computers, supercomputers and eventually GPU’s. But electronics design greatly diminished, and so did the demand for electronics designers.

Today, electronics designers still exist, but the demand is way down.

If you don’t like this analogy, consider the analogy of machinists (once in huge demand) and their replacement by CNC operators.

Programmers will be largely replaced, and salaries will come down: technology marches on and nobody is immune.