I'm not sure I agree with the doom and gloom here, but I do understand the sentiment. Especially after spending a few days writing something that I feel should've taken me 3 weeks. I do struggle with having a knee-jerk reaction to that change. I mean I just told my wife a couple nights ago that the future of programming feels shaky, even though I know that's not the reality right now. But she was in my office, so I showed her a bug, and then I asked Claude to fix it, and the bug was gone in 30 seconds, with regression tests.
So yes, AI writes code faster than I can, but it usually doesn't write better code. And you still need to know how to program to produce good code; it's very easy for Claude to write unmaintainable code, especially as it continues to write more code. You really have to put time into refactoring, using prior programming experience to know how to do so.
My current workflow is to prototype with Claude, and then refactor with Claude by giving clear instructions on what needs to be refactored and why. This works relatively well. But even then, at the end of the session, even the refactorings don't quite meet my high bar. So I hand-code towards the end. Maybe that last bit will go away? But it hasn't.
To be honest, code is too personal and artistic for me to fully give up control. I enjoy stressing the details, like doing a squint test on the code as I would with any art to check composition, or rewriting variables and classes and methods to hit the spot where I say "now that's beautiful code." Passing tests doesn't mean code good, or tests.
Thankfully my job allows me to spend time doing that. I think it's human to spend time doing that.
I'm not sure I agree with the doom and gloom here, but I do understand the sentiment. Especially after spending a few days writing something that I feel should've taken me 3 weeks. I do struggle with having a knee-jerk reaction to that change. I mean I just told my wife a couple nights ago that the future of programming feels shaky, even though I know that's not the reality right now. But she was in my office, so I showed her a bug, and then I asked Claude to fix it, and the bug was gone in 30 seconds, with regression tests.
So yes, AI writes code faster than I can, but it usually doesn't write better code. And you still need to know how to program to produce good code; it's very easy for Claude to write unmaintainable code, especially as it continues to write more code. You really have to put time into refactoring, using prior programming experience to know how to do so.
My current workflow is to prototype with Claude, and then refactor with Claude by giving clear instructions on what needs to be refactored and why. This works relatively well. But even then, at the end of the session, even the refactorings don't quite meet my high bar. So I hand-code towards the end. Maybe that last bit will go away? But it hasn't.
To be honest, code is too personal and artistic for me to fully give up control. I enjoy stressing the details, like doing a squint test on the code as I would with any art to check composition, or rewriting variables and classes and methods to hit the spot where I say "now that's beautiful code." Passing tests doesn't mean code good, or tests.
Thankfully my job allows me to spend time doing that. I think it's human to spend time doing that.