So here's a fun thing I've come to realize. Whatever gains I've been able to make by using AI to write code, I lose in the time it takes for it to write code. Stay with me...
When I was coding full-time and not managing, my output was not as voluminous as that of my peers. Now I know that it's because I have ADHD, but at the time, I figured that maybe I was lazy. I now better understand that the then-undiagnosed condition made it hard for me to concentrate on the work, especially if I wasn't able to get into the zone, which I also now understand to be what they call hyperfocus. See also: working in a cubicle office with countless distractions. While I like to think that I wrote quality stuff, it was definitely hard to write as much as others.
Fast forward to today, and I'm mostly writing code for fun on my own time. I've been getting a little deeper into it lately because the AI tools are like having a junior to mid-level developer pairing with you. If you give it the right directions and scope, something that definitely takes some time to learn, the outcomes are pretty OK. But there is a dark side to this as well, and it comes with the whirling icons or messages like "noodling." As text scrolls by and Claude does its thing, the desire to go do something else is overwhelming. The return time for usable code (assuming it compiles) can be at least 30 seconds, which is an eternity for someone with ADHD. I've estimated that in the course of five minutes, my inner dialogue may context shift at least a hundred times. Thirty seconds is a very big window.
I've been unknowingly developing coping skills for ADHD my entire life. When I got the diagnosis four years ago (also ASD), I learned about how neurodivergence forces people to find the shortcuts, the hacks, the compensation for having a brain that's wired differently. But this is the first time since then that I've found something totally new that I'm not used to. And it's crazy that I'm trying to compensate for a machine that's supposed to be helping me!
A lot of this is still right-sizing the work for the AI. When you ask it to take a big swing, it often gets stuff very wrong. But when you model entities up front, think about efficient ways to do stuff, and really design a solution, it's like giving the AI well-formed ideas and letting it figure out the glue. That's the boring part anyway, so knock yourself out, Claude. I do wonder how folks stay focused with the start-stop rhythm of using AI.
No comments yet.