I’m Done Using AI. “It made me lazy. It made me stop caring. It made me a worse programmer. It made me depressed. Because I stopped doing the hard work, I stopped learning, I stopped growing, I stopped being the one making the software.”
Advertise here with Carbon Ads
This site is made possible by member support. 💞
Big thanks to Arcustech for hosting the site and offering amazing tech support.
When you buy through links on kottke.org, I may earn an affiliate commission. Thanks for supporting the site!
kottke.org. home of fine hypertext products since 1998.
Beloved by 86.47% of the web.
Comments 16
thread
latest
popular
I have a hard time relating to things like this. My programming skills have improved more dramatically in the past 18 months than at any time in the past 20 years. I learn, interrogate, digest, expand, extrapolate more than ever before. I understand codebases much better, not less.
Steven Johnson has a good bit in the opening of this new Atlantic video about the whole cognitive offloading idea:
https://www.youtube.com/watch?v=Dmtv78xVf0M
Cognitive uploading is right.
I feel like I understand more things at a surface level, and fewer things at a deep level. The pathways from shallow to deep understanding are precisely what Claude's output intercepts.
Take your example of understanding codebases. I can understand a codebase much faster in the era of Claude. However, I am not building the skill set I would need to tear apart a codebase, and that skill set provides depth of understanding.
All that means is that you’re not able to take enough time to understand everything, especially given the likely much larger amount of code.
But the same could be said for moving from Assembler to Rust or Swift.
Claude & Co can be great tools for dissecting and understanding code.
I think there are (at least) two actually noble reasons people get into code. Some people more want to make amazing things and see their ideas emerge into daylight. Some people more enjoy the act of writing the code to solve discrete problems, more of a creative flow state. It's not a total binary. But people in the former category tend to be thriving lately under AI and people in the latter category are losing something.
I find myself in the latter and it's kind of a bummer. I can go off and write code for personal projects but, man, I used to come to work and do what I love all day.
This is a really great way of putting it, thank you!
I'm in the middle. I have (for over 50 years) loved programming. I get an idea of what to build, from myself, or from my employer, and then I start with nothing and design and build something. I'm there for coming up with an elegant architecture and design, building it, debugging it, tuning it. All the rest -- source control, code review (if I'm working with others), meetings -- is drudgery. And AI promises to take over everything but the drudgery.
I'm retired now, and still writing software, and exploring AI. I'm using local LLMs instead of foundation models, for many reasons. They are surprisingly good. I have a modest Mac Mini on my desk, which, with a local LLM, acts like an eager and talented intern. I have always been a backend programmer -- everything but the UI, so I am happy to turn the UI over to the AI.
The backend code that it generates is usually mostly correct. There are subtle bugs which I can see, but not only did the AI create the bug, the test it wrote didn't check for it! And the code is quite ugly. Now I assume that the frontend code has the same problems, but not being a frontend programmer, these problems are invisible to me, I'm vibe coding the frontend.
I'm getting things built faster than ever, but I'm less involved than I used to be. It's too easy to give up the important things I get from programming, and build an inferior system as a result.
I suspect that the disconnect is even worse with foundation models. They probably to a better job (on correctness, code quality), so its tempting to turn over even more work to them.
I coded in Objective C, Java and Swift and it paid my bills for 20+ years. So I understand how it feels to have something you invested so much into suddenly become commoditized.
But after spending the past year using AI to code daily, I've made peace with it. I'm a maker of software, not a writer of code. Turns out I enjoy seeing other people using my software more than writing code.
This is such a big identity and paradigm shift — but I think when so much of the industry is largely removed from end users (internal or external), these kinds of payoffs are also missing. The sub-unit ticket model has already made so many folks into cogs there isn’t much joy left but in the writing and making
100%. When you work in big tech, everything comes to you in tickets that are so atomized and so de-contextualized that the payoff of seeing customers using your software is practically non-existent.
Software engineering had become so disconnected from the original reason why I got into it that I was working like an LLM even before they showed up, and I was one of the lucky ones!
I've always thought they were 2 kinds of coders. Those who code because they like the intricacies of coding, language design, object or functional, etc.
And those who code to solve a problem.
The 1st kind will probably find this change a lot harder to adapt to, but I think it just means they have to find jobs where that's needed.
The 2nd kind I've found have loved this change, because it lets them solve problems faster, and in domains that they weren't as familiar with before.
Realistically though, most of the coding out there is the 2nd kind, most people aren't doing anything that revolutionary, we're all just building widgets for different domains.
IMHO this is the mentality of many software engineers who refuse to use AI.
Sometimes it's a valid; speed for the sake of speed isn't worth it. But doing things purposefully inefficiently in a way that reduces team collaboration is more harmful to an organization.
I've been trying to work with my developer for a year, but they're writing thousands of lines of code in Vim
This line seems to be the key to understanding the writer
I do not like this world where people are using AI to generate art, music, writing, and software
There are a lot of people that use coding as an expression of their creativity and I suspect the loss of that outlet to AI is what they're mourning.
I suspect this feeling is going to hit frontend developers more than "utilitarian" engineers such as myself (data engineer) because coding is just a (often repetitive) tool for me. Though I also think my skillset will be the easiest replicate via AI once companies get better at organizing and documenting their internal business practices.
I feel like so much of this comes down to whether or not language is thought.
Is it? I can't think without language. Or can I? Are we losing the tedious bits of language (I certainly don't miss the preamble on phone call) or are we losing our ability to think?
It is genuinely hard to tell.
So many comments here that I resonate with.
Personally, I've always written code to solve a problem, not just for the sake of code - though I respect the aesthetic value of "good code", whether that be demoscene works or realizing new research concepts.
AI is a force multiplier for me, letting me apply my domain knowledge (both on business and technical levels) to guide a more capable programmer to achieve my objectives. As a result I find it (mostly) deeply gratifying; the "problem solving" has evolved from something abstract into the ability to better define requirements - a skill that is directly transferable to my day job.
I feel like no one contends with the moral implications of building a dependency on technology that's wreaking so much havoc environmentally and economically. I'm reading Brian Merchant's Blood in the Machine and there's so many parallels. Maybe I'd have a better opinion on this if it weren't expressivly made to kill my career and the skills I spent years mastering.
I empathize. I think what I notice most about anything I work on that involves AI, whether it's research or thinking through nonprofit strategy, is that it's simply not memorable. Whereas I recently had lunch with my friend Crystal and we brainstormed some new nonprofit programs together over pita pockets. Do either of us have the kind of library AI does between us? No. But I retained so much more, and the work continued between us. My brain just doesn't value or hold onto the wall of text AI produces compared to a much simpler, organic conversation with a human collaborator.
If you feel like this comment goes against the grain of the community guidelines or is otherwise inappropriate, please let me know and I will take a look at it.
In order to comment or fave, you need to be a kottke.org member. Check out your membership options.
Note: If you are a member and tried to log in, it didn't work, and now you're stuck in a neverending login loop of death, try disabling any ad blockers or extensions. Or try logging out and then back in. Still having trouble? Email me!