These are my thoughts, written by me. I used AI only to sharpen up the grammar, punctuation, and formatting. Other posts on this blog that lean more heavily on AI are marked as such. Reference-style pieces tend to be more AI-assisted. Thought pieces like this one stay mine.
Lately, with AI work, I just have not been happy with anything I’ve produced. Are my expectations too high now that I can 10x anything I think of in a matter of time I couldn’t previously? Before agents were as good as they are now, I always had ideas, things to build, but my limiting factor was time. I’m a father of two, soon to be three, with a house full of animals and a life full of obligations. I didn’t have time to build all the things. But now, I can pop open Termius on my phone and get Claude Code or Codex to knock out something while I’m at baseball practice with the kids. Or I’ll spend an hour knocking out 80% of a project and run out of things to do, so I spend tokens and egregious amounts of time on what’s ultimately useless CI/CD things just because I can. I have religious CI practices on most of my side projects that won’t see the light of day, really for no reason other than I can. I’ve built so many stupid projects that solve imaginary problems and I don’t even know how they work or what they truly offer, just because I could.
I love being a software engineer. I loved honing my craft. I loved really getting to know a system, building genuine improvements, getting lost in the specific problem and the subsequent solution. There was something so magnificently satisfying about submitting a PR for work you genuinely owned end to end, that you polished up, tested, and documented. I loved the feeling of knowing that that ethereal “expert” position I chased was in my grasp, and that I was moving on to more and more complex ideas, systems, and projects.
I don’t get much of that with AI. Maybe I’m using it wrong, but I find it hard to really derive that same satisfaction anymore. Oftentimes in ceremony meetings, you’d catch someone saying, “Ok, I think we’re in the weeds a bit,” when a discussion started getting too technical. Engineers love being in the weeds, discussing the technical approaches, tradeoffs, etc. With AI, we’re not in the weeds anymore, and oftentimes being in the weeds was where the fun stuff was, at least for me. I’ve seen lots of posts and takes from {insert favorite tech bro here}, or just the average guy on Reddit, talking about basically how we’ve moved from writing code to getting more done, or how writing code was never the problem. Yeah, sure, I agree. But writing code was a skill, one that could be honed, and one that is just kinda… gone. Or at least, changed fundamentally, forever. Like, that’s a real skill that someone like myself has put years into learning, honing, perfecting, and it’s really gone, at least what it meant before.
Yes, the skills are transferable in this new era, and yes, you need to know how to code to really have AI be a force multiplier for you, so it’s absolutely worth learning and writing code by hand even in today’s world. But if you’re at an intermediate or senior spot in your career, the onus of your work now becomes quantitative rather than qualitative, and you’ll start leaning on AI so much instead of just doing what I referenced earlier: getting lost in the problem and solution, polishing, testing, owning.
I want to be satisfied in my work. I want to own my work. I want to get better at my career. I don’t want to be just shoving work out to get it done. I want to slow down, take my time, and really, really get quality software written. I want to get more done, just not at lightning speed. There was a sweet spot last year, at least for me, where I was hitting that. I deliberately took my time, I still owned things, I wrote a lot of code still myself. But agents can produce so much in so little time that now it’s hard to get into a groove with it.
I wrote a post a few weeks ago talking about being optimistic about this new era, and that I looked forward to the next generation of software, as it should be better than ever. I still look forward to this, but I’m having a hard time finding how to produce that next generation of software without hindering my own learning, pace, and career satisfaction. I’m just producing to produce right now. There are a lot of areas where producing to produce is fine. Like, oh, I need this quick little script to do some menial repetitive task, let me just AI that. But that has issues too over time. It compounds, because even in those little moments you learned something. “I spent 6 hours automating a 1 hour task” is a meme, but oftentimes you learned something useful in those 6 hours that you’d use later on, whether it was a lesson learned, a neat utility, a design pattern, whatever it was. Sometimes those little moments were huge. And with AI, it’s just gone… or shifted elsewhere that I’ve yet to find.
