✌️ Kenny Goff

No, I don't use AI

Photo of Kenny Goff
Kenny GoffJune 8, 2026

No, I don't use AI.

I don't use LLMs and I don't use Generative AI.

I don't use AI to code, write documents, summarize emails, search the web, create images, or prep my D&D sessions. I don't use AI as as my therapist, my companion, or my friend. I just simply don't use AI.

I am a programmer and I like programming. I like the craft. I like problem solving. I like working through weird tricky difficult and sometimes annoying problems to come out on the other side having learned something. I like the little dopamine kick I get when I figure something out. I am a curious person. I like learning. I also like designing. I like writing. I like drawing.

"Generative AI is catnip for the truly incurious"

‒ Ed Zitron, Better Offline episode "Big Tech Is The Enemy of Innovation"

I don't really want to have to write about AI - specifically LLMs and generative AI. But as a programmer who has dedicated years to a craft I enjoy surrounded by an industry that has been one-shot by automation porn, I feel like I gotta say: I don't fuckin' care about your LLM, I don't want to use them, and I'm simply just not going to.

The public sentiment towards generative AI seems to be mostly negative, especially for writing, music, art and video - the things most consider as creative fields. But I've also heard a lot of non-developers say "I hear programmers like it and it's good at generating code". As someone who's been programming since he was a teenager, professionally for over 10 years, and doesn't like or use LLMs - I want to offer my personal perspective.

Note: I'll mostly use "AI", "LLM", and "Generative AI" fairly interchangably. This is mostly just because the narratives of the last few years basically co-opted the very vague term of "Artificial Intelligence" to just mean Generative AI and chatbots. I know there are applications of machine learning and AI that are useful. I am mostly talking about LLMs and Generative AI here.

On using LLMs to generate code

I don't find writing code with an LLM enjoyable or much that useful.

  1. All the craft and joy I got out of the work is taken away
  2. The core skill of problem solving is replaced by the skill of wrangling a text-generation machine
  3. Non-deterministic output makes it difficult to make consistent and repeatable results
  4. Most things that I want to do multiple times I'm better off writing the automation myself to do it exactly how I want it every time
  5. Slot machine mechanics of reprompting is repetitive, inconsistent and ultimately slower than just doing it myself correctly the first time
  6. The quality of code that is output is mediocre at best and requires enough revision and review that it tends to just be faster to do it myself
  7. The code just looks weird sometimes, I can't place my finger on it but if you know you know - it does not look like a human wrote it and that makes it harder for me to read, parse, think about, and review.

There are also a few worrying things about this current environment we are in that feed on each other in a nasty cyclical way:

  1. LLMs are sychophantic and trend towards more and more code created - they are asked to write code so they will oblige
  2. An increase in code (especially the mediocre and bug-ridden kind) will result in more broken software and security vulnerabilities
  3. Companies are encouraging or outright requiring code be generated by an LLM pushes code output to its maximum
  4. Tokenmaxxing, token leaderboards, and enforced AI usage is a toxic and pointless "productivity" metric artificially increasing the slop code output and AI usage metrics for metrics sake
  5. Vibe-coders putting out code and products into the ecosystem they do not know how it works or if it's buggy or has security holes
  6. The software world is built on open source projects with underpaid, underappreciated, and overworked maintainers - these maintainers are getting bombarded with low quality slop contributions
  7. Companies claim AI can "do the job of a junior engineer" and use that as excuse to not hire junior engineers when the truth is a junior engineer's job is to learn and grow and become an experienced engineer through learning and growing which is something an LLM can never do
  8. Code generated by LLMs is either not being reviewed carefully because "eh, who fucking cares" or worse just being rubber stamp approved by a different LLM because literally no one fucking cares
  9. Prompt injection is a security nightmare that cannot be solved because its a feature of LLMs - they cannot distinguish between instructions and data
  10. Foundational model companies add all the new slopped out code they can get their hands on as training data for revised models
  11. Rinse and repeat

On using LLMs as a coding assistant or tool

If you avoided using LLMs to generate the actual code and instead exclusively used LLMs to do things line parse weird error outputs and scan PRs for errors then I guess they're fine as a pure assistant or extra layer of validation. If you ignore all the other environmental, economical, and ethical problems with LLMs of course ("hey chat what is 'willful ignorance'?").

I find that LLMs tend to be pretty good at finding a needle in a haystack, except it'll also find a button, a piece of lint, and a paper clip and say they're all needles with a big happy smile ("you're absolutely right this is a piece of lint"). So you really have to know enough yourself to know what is right and what is wrong. But to their credit that can still be useful if the needle is really hard to find.

I still personally prefer doing my own debugging and research because I find working through problems rewarding and fun. If thats not you or you are paged at 3am to fix an outage and need to parse tons of errors so you can go back to sleep I get it. It's just still not for me.

Responding to common arguments from AI boosters

"It's inevitable"

I tried to look at the price history of an SSD I bought two years ago to see what the price increased to in this current market. I clicked "Price history" and Amazon opened up a chat window with "Alexa for Shopping" and entered the text "Price history" as a chat prompt instead of just showing me a chart immediately. Why? They really really want to say their customers use AI.

It's inevitable in the same way it was inevitable the world would move to cryptocurrency and all art is now backed by NFTs. We all do that right?

I do not want to add yet another subscription and spend tokens to have a synthetic text extruding machine do my work for me. I disable AI summaries in my search engine of choice. I don't have Apple Intelligence enabled on my iPhone. I accidentally use AI when it's thrust upon me.

I don't want to be forced to use LLMs for work if I think the output quality isn't good and it just slows me down. I don't want a leaderboard of token use. I know how to do my job and I'm good at that job.

"You're a luddite"

Hell yeah I am. I'm proud to be called part of Ned Ludd's army and I take it as a compliment. Lets smash some machines.

I really love technology. I just hate the direction Big Tech and VC-backed Tech has gone. I hate the subscriptions, renting everything, dark patterns, and enshittification. I like owning my stuff, knowing how it works, having agency in how I use it and what I use, and being able to fix it. So fuck yeah, I am a luddite.

"Just a tool"

An LLM is "just a tool" in the same way stocking frames were "just a tool".

The output is shit at worst and mediocre at best, so even if its just a tool, its a pretty bad tool at the job it's claiming to do.

You effectively can't have an LLM without stolen content and high energy use to both train and for inference. They need tons of content for training and there's a reason model companies are using legally dubious scraping and stealing and are hoping to just pay some fines later or lobby for changes to how AI training data and output is treated with respect to copyright. The "Large" part of Large Language Models is a feature of them. They are also designed to be sycophantic. That's not an accident that is a deliberate design choice to get you using it more.

The underlying technology is mildly interesting to me. Natural language processing is a pretty cool technology. Are the outputs of LLMs worth the cost? I really don't think so.

Nothing is just a tool. The politics around it is real and it matters.

"You're going to be left behind"

I won't be left behind. I'll wait it out and I'll be on the other side of this with my cognitive skills and critical thinking intact.

"You're using it wrong"

I tried. I'm not trying to pass a purity test here. I was required to use it at my last employer. I tried to give it a shot.

I hated Cursor and tab completion. It made be duller as a programmer. I could feel myself reaching to tab instead of write code on personal projects where I didn't use any AI coding assistants. I hated that feeling and was so glad I had side projects to keep me sharp.

Using Claude Code and running multiple "agents" was a mildly fun toy to mess around with but my god the output quality of code was awful. And it cost so much in tokens to do such bad work. And this was way back before Anthropic and OpenAI started charging much more closely to real token costs.

"Use local models if you're concerned about ethicality, privacy, big tech, etc"

I would try local models. I tried to look into it. But the majority of them are not truly open source, just open weight. You cannot see what the training data is. I'm not inherently against LLMs, I just don't think the offerings and the industry that sprung up around them are worth my time.

Also if I have to build a crazy rig or a Mac Mini cluster in this economy ("in this economy!") just to run AI compute that's likely less good than the LLMs I've already tried... I'm just not sold.

If I ever explore using or building "AI" (a marketing term so broad it has basically no meaning anymore) it'll probably be models trained to do specific tasks well, not LLMs. LLMs if anything are a huge distraction and money pit taking away from the interesting parts of machine learning and "AI".

And again in a theoretical world where I could run a 100% ethically sourced farm-to-table local model... that doesn't fix the things I don't like about the usage and experience of LLMs or solve the fundamental problems with the technology.

"It will continue to get better", "Those problems will get ironed out", or "Oh you have to try this new latest generation of model"

Let me know if that happens! I'm happy to be proven wrong but the fundamental flaws of LLMs won't disappear with the wave of a magic wand. The text extrusion machine won't suddenly start thinking.

Closing

If you use AI because you are forced to at work I'm sorry. I understand the frustration and you're not alone. Do your best to survive even if it means using a thing you hate to meet performance reviews. Keep your skills sharp though so you can come out on the other end on top. If you can: push back, organize, and unionize.

I wrote this off the dome and the end of the day this was mostly a cathartic exercise for myself. I tried my best to tone down the snark from earlier drafts. Maybe other people find it useful or insightful but really I just wanted a place to point to so I don't have to explain myself over and over again.

Further reading, listening, and watching that I have found incredibly useful, educational, or just keeping me sane through this AI hype cycle: