AI killed my flow
I have started using AI in my coding workflow: with the constant mention of it and the endless stream of ground-breaking improvements, it was only a matter of time before I plugged it in and gave it a whirl.
Turns out, it's not that fun. It's incredible, powerful, insightful, almost always (but-never-quite) right, but it's not fun.
I spent years building my craft like anyone who decides to pursue a career in something, whatever it may be, and I spent years choosing my tools and reading thousands of lines of code to understand better how good software is built (which doesn't mean I do).
But it's been really fun, especially trying out the arsenal of tools to help us navigate the fascinating world of computers. Learning Vim in university was amazing, and switching to Emacs ever since as made my job a happy place even at harder times. You know why? Because they put me in the driver seat, and I like driving. And whatever IDE you use, if GEdit/Sublime/TextMate/name your favourite antique text editor, was your thing, I hope you found the same joy. Or VSCode, I don't actually care. This isn't really about editors, but if you know you know.
Anyway, I liked the flow it put me in and I liked how sometimes I could lock in for hours and eat at 10PM because my brain was wired and I couldn't step away from the code. I'd also just gotten to a point where, 15 years later, there are no major roadblocks when I code things – I've learned to debug better, I clone libraries and try to understand the code when I get stuck with them, I do better note-taking, you know, the whole shebang (sparing the bash joke here).
Now AI works it harder, makes it better, does it faster, makes us stronger 🎵. There is no doubt, no question about who is a better coder, and who will be in 2, 3 years time (AI). But this new way of working, this powerless state of watching an AI stream some code via The Endless Text Cascade, getting it almost right, and then having to test/review what it's done, is just not as fun.
Actually having to review/debug hundred of new lines at once is excruciating, and this is where a decade and a half of programming comes in handy to understand what's gone wrong: I'm guessing the more junior programmers who are using AI to shotgun all of their problems are going to experience some form of anxiety having to judge or fix the result without an actual grip of the system.
Hell I also get myself into a right mess when I use AI sometimes. I'm
not one of these people who commit with git -am "fix"
. I rebase a
lot, I try and put things where they belong, whether we're talking
about code or commits. I can't function otherwise: computing is so
hard, you gotta understand what you're doing.
I guess I just have to find the right involvement for everybody. I want insights from AI, but letting it do all the work removes all the fun I had doing mine.