I'm not who you are responding to, but I do similar things with AI.
Think about it this way. We all make decisions like this pretty much every day, but I am especially careful with them in my personal life where time is limited and sacred: "What amount of time or money (X) am I willing to spend to get something (Y)?"
There have been many times that X has been "too much," until I later discover some new tool, library, or technique (or simply a price drop) that reduces X below the threshold of pulling the trigger. AI is that new tool for a lot of people and contexts.
If my barrier to some cool new one-off home automation feature is something like, "I would need to know Ruby but I don't know it and don't have time or desire to learn it," then I can have an LLM do the heavy lifting in a tiny fraction of the time it would take me to learn. Of course the feature needs to be something straightforward enough for the LLM to handle, and you have to be able to test it. And it goes without saying that since I can't properly review the code, I wouldn't use it for something that could cause a lot of damage or a security issue. But there are lots of tools/areas where that is not applicable. (Not all code needs to be bullet-proof and in reality, almost none of it is, even when it should be.)
Think about it this way. We all make decisions like this pretty much every day, but I am especially careful with them in my personal life where time is limited and sacred: "What amount of time or money (X) am I willing to spend to get something (Y)?"
There have been many times that X has been "too much," until I later discover some new tool, library, or technique (or simply a price drop) that reduces X below the threshold of pulling the trigger. AI is that new tool for a lot of people and contexts.
If my barrier to some cool new one-off home automation feature is something like, "I would need to know Ruby but I don't know it and don't have time or desire to learn it," then I can have an LLM do the heavy lifting in a tiny fraction of the time it would take me to learn. Of course the feature needs to be something straightforward enough for the LLM to handle, and you have to be able to test it. And it goes without saying that since I can't properly review the code, I wouldn't use it for something that could cause a lot of damage or a security issue. But there are lots of tools/areas where that is not applicable. (Not all code needs to be bullet-proof and in reality, almost none of it is, even when it should be.)