Yeah he decides “AI is good actually for coding” but the same reasons apply there.
(Note: Non-coders prototyping is a different use case than professional coding - I am talking about professional coding here)
LLM code lets people shit out systems with no intent. And those systems are a waste of time. They just feel “90% there” for people who never had to finish a project in their life.
In the small scale, I get PRs of 200 lines of JS from someone who felt super productive because Claude wrote it for them. What did the JS do? Replicate the CSS “transition” property badly. They could have written one line instead. And gotten a bug-free, efficient, readable and maintainable version of what they were trying to do instead.
In the large scale, you get thousands of lines of RPC middleware instead of someone saying “hey at this point, should we move this responsibility from module A to module B and get rid of a lot of code?”
And I refuse to be on the defense of luddite or hater like the author does. Because I have never heard that claim from anyone who is capable of actually shipping stuff better or faster than me.
This has been going on a lot longer than AI, where the programmer (in this case the AI) isn’t aware that a function that does what you need already exists, it goes making it from scratch. It ends up slower and less stable than the proven function, but they get it to run, so to them it’s a success.
When I started to get a grasp on programming, I realized early on that I’m rarely solving novel problems and I should seek out to see how it was solved by others before I even begin to write my own approach. I usually find someone has done all the work for me and theirs works better than mine ever would.
Yeah he decides “AI is good actually for coding” but the same reasons apply there.
(Note: Non-coders prototyping is a different use case than professional coding - I am talking about professional coding here)
LLM code lets people shit out systems with no intent. And those systems are a waste of time. They just feel “90% there” for people who never had to finish a project in their life.
In the small scale, I get PRs of 200 lines of JS from someone who felt super productive because Claude wrote it for them. What did the JS do? Replicate the CSS “transition” property badly. They could have written one line instead. And gotten a bug-free, efficient, readable and maintainable version of what they were trying to do instead.
In the large scale, you get thousands of lines of RPC middleware instead of someone saying “hey at this point, should we move this responsibility from module A to module B and get rid of a lot of code?”
And I refuse to be on the defense of luddite or hater like the author does. Because I have never heard that claim from anyone who is capable of actually shipping stuff better or faster than me.
This has been going on a lot longer than AI, where the programmer (in this case the AI) isn’t aware that a function that does what you need already exists, it goes making it from scratch. It ends up slower and less stable than the proven function, but they get it to run, so to them it’s a success.
When I started to get a grasp on programming, I realized early on that I’m rarely solving novel problems and I should seek out to see how it was solved by others before I even begin to write my own approach. I usually find someone has done all the work for me and theirs works better than mine ever would.