ML-Enhanced Code Completion Improves Developer Productivity (2022)

"cigarettes do not have adverse health effects"

- cigarette industry

I don't have any data for this - so it's just a conjecture - but I think ML-Enhanced Code Completion and other AI developer tools levels the playing field for anyone who enjoys pair programming but doesn't have someone to participate, or gets analysis paralysis or blank page syndrome. I think this sort of thing is why you are getting the wide spectrum of feedback about how good AI coding tools are. For some, who already had smooth transition into flow it's not such a big deal but for others they might be experiencing a neo-flow like state for the first time possibly.

As always, the results depend on the user. For a small toy project, the r results are usually incredible. For a large mature production app - the results are borderline useless.

Every year the ai gets significantly better.

I suppose the hot take on this is something like: there are detractors, but I think for the majority of software engineers - and certainly those using premium plans of the various available providers - this is not news.

Being perhaps a little bit more thoughtful: 2 - 3 years after the beginning of widespread adoption of ML - specifically transformer - assisted software development, having some actual numbers that demonstrate its value and can help to press the case for its adoption is in fact useful.