I will not be using LLMs in my coding

This is a continuation to the past two blog posts I have made about coding and AI. After quite a lot of soul-searching, I have come to the conclusion that I will actively not be using any LLMs or AI agents in my professional work. It's been quite a rollercoaster this year, to say the least. There have been excellent articles and opinion pieces that have resonated with me and my feelings around AI usage. In this blog post I will list my biggest dealbreakers on why I won't be using LLMs in my work. Recommended reading I highly encourage everyone to read the essay 'Why I Resist Generative "AI"' from Esther Weidauer. They managed to articulate so much better than I ever could what's bugging me about the current enthusiasm of using AI in coding. Mind you, I do not 100% agree with every point in the article, but I definitely share the main sentiment and concerns. I especially like the section about how people argue that "LLMs are just a tool", but really, it is not, it is a SERVICE, where you outsource your work to the whims of a (usually US-based) third-party company. I know Machine Learning, trust me Before anyone calls me a luddite or a AI-skeptic, I want you to understand that I know Machine Learning. I have built production-grade ML implementations for several clients in my career for the past 10 years or so, like CNN-based image recognition models in Keras, air quality prediction models in Tensorflow and more. I have also dabbled with locally running LLMs and image diffusion models in my homelab, and have even contributed back in Github. LLMs and transformers are a cool and exciting technology, but the current hype and craze for them is frankly overblown and hazardous to people's work and health. My dealbreakers These are the dealbreakers, for me personally, on why I actively try not using LLMs in any of my professional work. Cognitive impact I've heard of the "LLM brainfog" from my fellow programmers all this year. And it's no wonder, there have been several studies confirming this: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers (from MICROSOFT!) Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task Large-Language Models as a Cognitive Virus, how LLMs make you more and more reliant on using them for all tasks When I personally have used LLMs for several hours, I often feel disoriented and confused, and I feel like this is not a healthy thing for me to participate in. The quality of the output is not that great I feel that LLMs and their overtly verbose output trick less knowledgeable programmers of the quality of their output. When Claude spits out refactored code with tons of comments, documenting each method and variable, and tests for everything as well, it looks impressive, at a first glance. But if you dig deeper, it's sometimes just complete bullshit. I have had several cases at my day job where I've had to review Claude-generated code for bugs that could have been solved more elegantly with just a couple of lines of dense schema changes and migrations, but Claude had decided to build something completely new, utilizing the latest hooks and methods in the framework, outputting several tests for a single field etc. LLMs expose the literacy gap between great and mediocre coders. If you can't see how bad the generated code is, LLMs will only make you not see it faster. Maintainability is a nightmare Closely related to quality of code output, I've begun to see that the maintainability issue of LLM generated code is starting to creep in. If all of the code is written by agents, then at some point no human will understand what's going on, and that's not something I am comfortable with. Congratulations, you have let OpenAI/Anthropic/Elon Musk dictate what your code does and why. I like writing code I have written about this several times in the past. I just simply love writing code. It's part of my process of problem-solving. I constantly have small epiphanies when I write code; what's the optimal way to iterate over the content, where should this data be stored in the hierarchy of the application etc. This also applies to reading other people's code. I want to understand how the human behind the code thinks, plans and executes. What's their motivation, their goal, their field of expertise. AI has zero of this nuance in its output. Environmental aspect I probably don't need to say too much about this. Everyone has heard the news about new AI datacenters and their electricity usage (sometimes even using new fossil fuel sources!). Here in Finland, the news were briefly dominated by Google, on how they cut down hundreds of hectares of forest, without a permit, to build a datacenter. The "datacenter mania" is clouding people's minds and that's bad for the planet. Politics This is the hardest point for me to articulate. But I just personally feel squeamish about having my work being done by rather skeezy companies like OpenAI, Anthropic, or, god forbid, Cursor/SpaceX. Also, all the big AI providers are US-based, so there is definitely GDPR aspect to everything as well, me being an EU citizen. Everyone seems to have forgotten all about this in their big AI craze of 2026. Even though if a company SAYS the data stays in the EU region, remember that if they are a company based in the United States, CLOUD Act grants federal law enforcement authority to receive data stored anywhere in the world. There's been some talk online how "AI-assisted coding is tool of the elite", to lock you into a dependency of a commercial software in your line of work and livelihood. I feel it's been a bit exaggerated, but I can't help but feel like there's a nugget of truth somewhere in there. There's also positives, yes Now, there definitely have been positive moments when I have used AI tools in my coding work. For example, if I'm completely stuck somewhere in implementation, I can articulate a simple question to an AI agent and get somewhat of a coherent plan of execution forward. But in the end, the help I get from it I feel like is just not worth it, due to all of the above-mentioned reasons. AI is not inevitable I love how Esther Weidauer summarizes this argument: "AI is here to stay." OK. So is heroin. And I wouldn’t recommend that to anyone either. "AI is here to stay." OK. So is heroin. And I wouldn’t recommend that to anyone either. I keep hearing from all around me how inevitable AI is. And I would like to push back a little on this. Yes, AI-assisted coding is not going away. No technology can just disappear overnight. BUT it does not mean that you have to use it. With these subsidized token prices currently, I bet were seeing less and less AI usage encouragement in companies in the future. Even if some technology or tool sticks around, it does not necessarily mean its good idea to work with it. Conclusion This is probably my last post on this topic, at least for a while. I'm quite done researching and hearing about AI agents, harnesses and whatever else. I'm going back (or did I ever leave?) to old school coding. I tried writing this in a way where I share my personal feelings about this issue. I do not want to force or change anyone's opinion on AI-assisted coding. If you are happy with your current AI-assisted setup, then go for it. But these are my personal pain points that I have regarding its usage in the software world. Maybe I'm wrong, maybe I'm right, who knows. Frankly, at this point, I do not care, I just want to write my code in peace, with my human hands.