Blog entries, Hacker News entries, and such as this one are starting to appear regularly.
https://crawshaw.io/blog/programming-with-llms
This specific one was noted on ArsTechnica, so expect it to be seen by many.
While I agree with most of what is noted, one key distinction or breakout that may not be known by many but is quite valuable is for Chat-driven programming, in that I see this as two different ways.
- Using a prompt to create the entire code and then refactor the code using just the prompt.
- Using a canvas, selecting a specific part of the code, then entering a prompt. This method is often more productive as the prompts are more often understood correctly, the code changes are more accurate, and you don’t have to hunt and peck to know what part of the code was changed.
Additionally using AI to create documentation, especially with the o1 models is at times even more useful than creating the code.
So in essence this is an answer looking for a question that I don’t know if many know to ask. 
David, thanks for the blog entry. Nice to see such information being passed along.