Listen to this: “I don’t use AI for programming, because I want to learn and get better at programming and if I have an LLM generate code for me, I’m not learning anything.”
I think your stance is reasonable precisely *because* you have "written enough web apps in my life to know that I’d mostly end up being frustrated after trying to figure out how to get to the URL query params".
For others who are relatively beginner-ish, they might be better off writing the programs themselves to learn the skills needed to program than just their work done. Even in an LLM-dominant world, it's going to be useful to be a good programmer to be able to debug AI-generated code.
Fundamentally, I think there are two different modes you can be operating under - learning mode vs. getting things done mode - depending on the situation. Not trying to shill here, but since this is exactly the premise of a recent blog of mine, and I want to avoid duplicating the arguments, here is the link: https://devanshshah.dev/posts/learning-vs-getting-things-done/
I am one of the people you paraphrased at the start but I also agree with you. I have started using Claude recently and find it helpful for small things where I want the end result more than the journey to get there. Those things are mostly limited to quick personal projects that are meant to solve a problem. I don’t use AI for assistance with code I write for work because I want to be confident that I understand all of it as I write it.
When I tried copilot I found myself not taking as much time as I should to read and understand what it generated for me. Not understanding the code you write and the purpose behind it leads to more bugs and more difficulty fixing those bugs.
AI is here to stay but it’s just another tool. Tools, when used properly, are extremely helpful but some people still enjoy doing things “the old fashioned way.” It feels strange to think that writing code with all the modern tools except AI could be considered old fashioned but I think many people would agree with that.
It takes switching your mindset a little bit to move problems that are in the “I could make a little utility for this but I would never bother because it would take too long” pile to the “chatgpt can spit it out for me in 1 minute” pile.
I have lots of random utilities at work including userscripts that make tiny tiny improvements that would have never been worth it to write by hand. It’s about spotting these opportunities now.
fundamentally disagree
This is like Napster drama. They were not in CD printing business, and we’re not in line writing business. A program is not LoC.
I think I agree, yeah
That small feeling of satisfaction that comes from solving throwaway problems might be my favourite part of writing code
I think your stance is reasonable precisely *because* you have "written enough web apps in my life to know that I’d mostly end up being frustrated after trying to figure out how to get to the URL query params".
For others who are relatively beginner-ish, they might be better off writing the programs themselves to learn the skills needed to program than just their work done. Even in an LLM-dominant world, it's going to be useful to be a good programmer to be able to debug AI-generated code.
Fundamentally, I think there are two different modes you can be operating under - learning mode vs. getting things done mode - depending on the situation. Not trying to shill here, but since this is exactly the premise of a recent blog of mine, and I want to avoid duplicating the arguments, here is the link: https://devanshshah.dev/posts/learning-vs-getting-things-done/
I am one of the people you paraphrased at the start but I also agree with you. I have started using Claude recently and find it helpful for small things where I want the end result more than the journey to get there. Those things are mostly limited to quick personal projects that are meant to solve a problem. I don’t use AI for assistance with code I write for work because I want to be confident that I understand all of it as I write it.
When I tried copilot I found myself not taking as much time as I should to read and understand what it generated for me. Not understanding the code you write and the purpose behind it leads to more bugs and more difficulty fixing those bugs.
AI is here to stay but it’s just another tool. Tools, when used properly, are extremely helpful but some people still enjoy doing things “the old fashioned way.” It feels strange to think that writing code with all the modern tools except AI could be considered old fashioned but I think many people would agree with that.
That seems like a reasonable stance, yeah.
It takes switching your mindset a little bit to move problems that are in the “I could make a little utility for this but I would never bother because it would take too long” pile to the “chatgpt can spit it out for me in 1 minute” pile.
I have lots of random utilities at work including userscripts that make tiny tiny improvements that would have never been worth it to write by hand. It’s about spotting these opportunities now.
Was that 'until it goes click' a Big Lebowski reference?
Absolutely