Genuine question, how can anyone do everything you say and still save time using AI? If you have to check everything it does to that level of detail, which I agree you have to, then where are you saving time? The only thing it “saves” me when I try to use it is the typing, which is far from the bottleneck. And in exchange I have to do the prompt and an extra very deep code review of the code I didn’t type which takes way longer for any non trivial change.
I also thing you are mising the point. Reviewing my own code or code written by a person is orders of magnitude easier than reviewing AI code. The mistakes a human makes are usually simple off by one errors and similar, or missing understanding of edge cases which are easy to spot in a review. The mistakes made by AI are usually a small random variation of an okish solution which are incredibly easy to miss given that the brain autofixes on the fly, just like trying to find small typos in a wall of text.
It’s also wrong to pretend that AI doesn’t promote the kind of laziness you are critiscising, specially because people are expecting huge productivity gains.
In my experience AI is barely useful as a SO alternative, and even then is hit and miss for anything not trivial. For example, I tried to use it when starting a new KMP project in which I wanted to share code between jvm and android. I figured that this being a boilerplate configuration task it could accomplish it, but it just produced non working gradle files while trying to create a nonsensical project structure. To me, everytime I try to use it on anything but the most common technologies and the most standard code structure it fails misserably. So again, I’m curious where do you see it saves you time?
Genuine question, how can anyone do everything you say and still save time using AI? If you have to check everything it does to that level of detail, which I agree you have to, then where are you saving time? The only thing it “saves” me when I try to use it is the typing, which is far from the bottleneck. And in exchange I have to do the prompt and an extra very deep code review of the code I didn’t type which takes way longer for any non trivial change.
I also thing you are mising the point. Reviewing my own code or code written by a person is orders of magnitude easier than reviewing AI code. The mistakes a human makes are usually simple off by one errors and similar, or missing understanding of edge cases which are easy to spot in a review. The mistakes made by AI are usually a small random variation of an okish solution which are incredibly easy to miss given that the brain autofixes on the fly, just like trying to find small typos in a wall of text.
It’s also wrong to pretend that AI doesn’t promote the kind of laziness you are critiscising, specially because people are expecting huge productivity gains.
In my experience AI is barely useful as a SO alternative, and even then is hit and miss for anything not trivial. For example, I tried to use it when starting a new KMP project in which I wanted to share code between jvm and android. I figured that this being a boilerplate configuration task it could accomplish it, but it just produced non working gradle files while trying to create a nonsensical project structure. To me, everytime I try to use it on anything but the most common technologies and the most standard code structure it fails misserably. So again, I’m curious where do you see it saves you time?