Table of Contents
I have been a sysadmin (systems administrator) since my teenage days when I made my first forays into learning how to use Linux. I learned the craft by hosting things that were useful to me in some way. I learned how to set up and maintain a web server because I wanted to host my own WordPress blog which eventually became this blog. Along that road I also learned how to set up and maintain a MySQL server, and how to run PHP applications on said server.
Later, I would use those skills to maintain services for other people. I learned how to set up and maintain a IRC server because I had friends who wanted to start our own IRC network. This was in 2011. That IRC network is still going. It is the longest continuously running service I have maintained. I am proud of it.
These days, sysadmin is both hobby and dayjob. I have been a professional sysadmin for a decade or so, a hobbyist for longer than that. That hobbyist aspect is still running strong, as an example I maintain the infrastructure for Ten Forward and have since its inception in April 2017, taking on the double role of sysadmin and community moderator.
I provide all this context to elucidate that systems administration is a core part of who I am. It is something I have a passion for and it is something I care about quite deeply and have for a long time. So when on the morning of February 4th, 2026 I saw a Ars Technica blog post written by Lee Hutchinson in my RSS feed reader with the title, “So yeah, I vibe-coded a log colorizer—and I feel good about it“, I knew I was gonna have another angry morning. So, I went on a rant about it on my personal fediverse account.
This blog post is me turning that rant into something a bit longer, better written, and something that is less ephemeral than posts on the fediverse usually are. All quotes in this post are from the text of the Ars Technica blog post unless specified otherwise.
Let’s get into it.
You Set The Planet On Fire and All You Got Was A Log Colorizer
Like Lee, I am also a sysadmin who is not a programmer. The most complicated thing I have coded was a Python script parsing a PDF and interacting with the Twitter API and that was many years ago. These days, I occasionally write simple Bash scripts to automate something at my dayjob or on one of my personal servers. I am not counting the YAML I write in the form of Ansible playbooks as programming though there is an argument to be made that it is.
Lee’s blog post starts off reasonable.
I can’t code.
[..]
Oh, sure, I can “code.” That is, I can flail my way through a block of (relatively simple) pseudocode and follow the flow. I have a reasonably technical layperson’s understanding of conditionals and loops, and of when one might use a variable versus a constant. On a good day, I could probably even tell you what a “pointer” is.But pulling all that knowledge together and synthesizing a working application any more complex than “hello world”? I am not that guy. And at this point, I’ve lost the neuroplasticity and the motivation (if I ever had either) to become that guy.
Thanks to AI, though, what has been true for my whole life need not be true anymore. Perhaps, like my colleague Benj Edwards, I can whistle up an LLM or two and tackle the creaky pile of “it’d be neat if I had a program that would do X” projects without being publicly excoriated on StackOverflow by apex predator geeks for daring to sully their holy temple of knowledge with my dirty, stupid, off-topic, already-answered questions.
I understand some of this frustration. I have long since grappled with my inability to truly grasp programming as a skill and a craft. I have dealt with the toxic nature of various help forums like Stack Overflow. Trust me when I say that Stack Overflow has nothing on the toxicity of the gentoo-users mailing list a decade ago or the various IRC channels one could ask for programming help. I get it.
Unlike Lee, I have made peace with the fact that I am not a programmer and cannot seem to find any intrinsic joy in the act of programming. For me, it is not a matter of neuroplasticity or motivation, it is just not something I am interested and such is life. Not everyone needs or wants to be a programmer. Its fine. I am content to be a system administrator and leave the programming to the programmers.
Even if programming was my white whale, I certainly would not be “talking” to an LLM about it. And that’s where this whole post goes off the rails.
Before I talk about some of the finer points of the blog post, I am going to quote from the conclusion.
The log colorizer is not the first nor the last bit of vibe coding I’ve indulged in. While I’m not as prolific as Benj, over the past couple of months, I’ve turned LLMs loose on a stack of coding tasks that needed doing but that I couldn’t do myself—often in direct contravention of my own advice above about being careful to use them only in areas where you already have some competence. I’ve had the thing make small WordPress PHP plugins, regexes, bash scripts, and my current crowning achievement: a save editor for an old MS-DOS game (in both Python and Swift, no less!) And I had fun doing these things, even as entire vast swaths of rainforest were lit on fire to power my agentic adventures.
Yes, Lee really did write that last sentence. The first time read this article, I thought I had misread it, that I had hallucinated this sentence. But, no, it is something Lee not only wrote but published on a high traffic site for people to see. If this is the conclusion you come to then you need to stop writing, go stare at a wall and contemplate the choices that brought you to this completely unhinged, abhorrent conclusion.
It is quite hard to describe the sheer disgust I felt reading that last sentence. A couple people on the fediverse didn’t believe it was a direct quote. Yeah, part of me still can’t believe someone wrote this and published it on the internet to see. It is not only something I would never write or think but I wouldn’t even say it to my closest friends because they would rightfully say, “hey what the fuck dude?”.
The sentiment on display in that quote is why we are in this goddamned mess to begin with. The extremely short term thinking, the absurdly selfish desire to solve a personal problem at the cost of the entire planet. All of this is emblematic of how LLM bros and the vulture capitalists pumping this bubble think. Who cares if the planet is ruined? You had fun *and* you got a log colorizer out of it. All of this over a goddamned log colorizer.
The Good for Nothing Nature of LLMs
LLMs can be fantastic if you’re using them to do something that you mostly understand. If you’re familiar enough with a problem space to understand the common approaches used to solve it, and you know the subject area well enough to spot the inevitable LLM hallucinations and confabulations, and you understand the task at hand well enough to steer the LLM away from dead-ends and to stop it from re-inventing the wheel, and you have the means to confirm the LLM’s output, then these tools are, frankly, kind of amazing.
But the moment you step outside of your area of specialization and begin using them for tasks you don’t mostly understand, or if you’re not familiar enough with the problem to spot bad solutions, or if you can’t check its output, then oh, dear reader, may God have mercy on your soul. And on your poor project, because it’s going to be a mess.
These tools as they exist today can help you if you already have competence. They cannot give you that competence. At best, they can give you a dangerous illusion of mastery; at worst, well, who even knows? Lost data, leaked PII, wasted time, possible legal exposure if the project is big enough—the “worst” list goes on and on!
*takes a few deep breaths*
What the actual fuck are LLMs good for? If you have to be familiar enough with a problem space, if you know subject area well enough to see the hallucinations and you understand the task enough to steer the LLM away from dead ends, what the actual fuck is the LLM actually doing for you then? If you meet all those conditions why are you not solving the problem without the help of a LLM?
As far as I can see, it is all downside with no upside. Any potential upside is covered under a nightmarish layer after layer of hallucinations, incorrect assumptions and a machine that doesn’t actually *know* anything. So again, what in goddess’ fucking name is the point of using a LLM for this?
The Good Sysadmin
I am a better sysadmin than I was before agentic coding because now I can solve problems myself that I would have previously needed to hand off to someone else.
Let’s interrogate this sentiment. Why do you feel the need to solve every problem yourself? What is wrong with having to hand off the problem to someone else? No, using an LLM didn’t make you a better sysadmin. At best you are the same level of sysadmin you were before you did this. At worst, you regressed your problem solving skills.
Being a good sysadmin means knowing the limits of one’s skills and knowledge. To be able to go, “okay, I can’t figure this out, lemme ask another person for help”. That is not something to be ashamed of or a mark of incompetence. Did you ask someone else for help? Another sysadmin? Maybe a programmer friend to help with the colorizer? Go on IRC, forums, hell even Stack Overflow? A log colorizer is a solved problem.
I deal with cache invalidation issues in relation to WordPress sites and Cloudflare at my day job, and there are other people like me who could have helped you with the underlying cache invalidation problem you were trying to figure out. Instead, you went about it the most roundabout, inefficient way possible and when the plagiarism machine gave you a colorizer, it was *you* that used your experience and knowledge to find the issue, not the LLM.
Have you ever been stuck troubleshooting an intermittent issue? Something doesn’t work, you make a change, it suddenly starts working, then despite making no further changes, it randomly breaks again.
The process makes you question basic assumptions, like, “Do I actually know how to use a computer?” You feel like you might be actually-for-real losing your mind. The final stage of this process is the all-consuming death spiral, where you start asking stuff like, “Do I need to troubleshoot my troubleshooting methods? Is my server even working? Is the simulation we’re all living in finally breaking down and reality itself is toying with me?!”
In this case, I couldn’t reproduce the problem behavior on demand, no matter how many tests I tried. I couldn’t see any narrow, definable commonalities between days where things worked fine and days where things broke.
I run into situations like this all the time at work and what I do is ask my coworkers to take a look and see if they have any ideas. It is in fact the first thing I do before I even consult a search engine because I find that asking my coworkers often gives me better results than asking a search engine because humans are flexible and can use their experience and knowledge to come up with ideas and solutions to problems in ways that a search engine cannot.
When I am deep into the weeds of solving a issue like this, I get tunnel-visioned into one particular perspective on the issue and getting past that involves other people with different perspectives and specializations coming in and going – “have you looked at $X? this sounds like a issue caused by $Y” and then going “oh huh I hadn’t considered that.”
I usually end up learning something from such an experience and that in turn makes me a better sysadmin than I was before I asked someone for help.
The Lone Wolf Dies In The Cold
I sense the undercurrent of a very individualist perspective here. A lone wolf sysadmin against the Big Bad Problem solving the problem at great cost to oneself and to the world around them. It is honestly very American but this isn’t just an American thing, it is way more endemic than that in IT circles.
The only thing LLMs are good at is feeding one’s ego, it will give sycophancy when what you needed was a helping hand and that’s where the con finds its mark. A good con works because it feeds on some sort of emotional desire and in this case the LLM fed on your desire to solve this issue all on your lonesome.
Fundamentally, I don’t think that system administration as a craft and as a practice is a solitary activity. Systems are often complex with multiple moving parts. You cannot know every detail of every little gear. You will specialize in a small subset of the system that you understand very well and others you only understand in abstract or not at all.
That is where system administration becomes a communal craft. In fact it always has been. As an example – reading and understanding documentation for a system you maintain is part and parcel of being a good sysadmin. A lot of the time the documentation will have been written by others, the creators/previous maintainers of the system. By reading their documentation you are making use of a communal resource, it is by definition leaning on others.
Documentation is only one small aspect of the communal nature of systems administration. Asking people for help directly is another. Your peers will have different sets of skills, knowledge, and experience that will not only help you solve your issue but will in turn strengthen you and make you a better sysadmin. You can then pay it forward by helping someone else. “Better Together” is a pithy saying but it very much applies here.
Using a LLM for such purposes cuts out these vital aspects from your chosen craft. It leaves you disconnected from your community and talking to a machine that does not actually care about anything and as such cannot make you better at anything.
You cannot vibe code your way into becoming a better sysadmin. Or better anything else for that matter.
Thank you for reading this blog post. Please consider supporting my writing work directly through Stripe or Patreon. Additionally, please share this post on your socials and/or with your friends. It really helps! If you have any thoughts/feedback/thoughtful criticism to share, feel free to use the handy contact me page to send that to me.