Keep in mind that this will be stats for the HN crowd, not the industry overall.

Also, I find it far more powerful for planning and solo brainstorming than for the coding. For every piece of code that it speeds up, it also means that much more code that is not baked into my head, so my debugging and conceptualization of the codebase is weaker. And that isn't even taking into account the need to check an AI's work.

But when I need to do something utterly unlike any of my previous work, instead of digging in for a day or two just to conceptualize a novel feature, I can get a plan of how to approach and architect the feature in a few minutes. And the quality doesn't really matter at that point - if it is good, I can run with it. If it is bad, it gives me a baseline to react to and make a better plan, which also speeds me up.

I use it for bite-sized things. Extremely useful. I've been able to build things I would never have attempted before. (Engineer by profession, programming little scripts here and there). It gets me over the "activation energy" hump.

Some of the influencers are so annoying in pushing stuff that doesn't work that it seems appropriate to not try stuff just to spite them, but the problem with LLM is that things that didn't work last year have started working and this keeps repeating. In many cases they aren't really wrong, they are just early (although maybe it's the stock picking hindsight bias as to whether they are good prognosticators or not).

Biased poll options. Please redo. Suggest Likert scales.