Conversation

On using natural language tooling to write software:

If programmers do it to save time on boilerplate, that can be helpful, even though arguably that seems the language and libraries don't have the appropriate levels of abstraction.

If programmers do it to learn from examples and quickly catch up on stuff they haven't done before, that can be great.

I have friends who do either or both of the above.

If people who can't program use a statistical engine to generate code they don't understand and can't validate, and they trust the results blindly because they also don't understand QA and testing, we're entering an era of more and weirder bugs than ever, on all levels of abstraction from high-level logic down to API calls.

On the positive side, for the few programmers left who can write formally verified programs ... or who can write things like AIs and system libraries, maybe their salaries will be worth tenfold.

The sustainability of that crowd is already in question though. They're on average only slightly below 20 years older than hey were 20 years ago.