Vibe Coding

Wed, 07 Oct 2026 15:59:45 +0000

AI is a very polarising topic right now, and I’m going to write down some of my views, which will further solidify why AI is so damn polarising.

Money grabbing hype machine

First of all, the level of hype in the media is insane, and as far as I can tell it’s all because AI is turning over more venture capital money than most countries’ GDP.

Take for example the recent Rogue AI stories, which are complete BS. There is no such thing as a rogue AI, there is only humans using AI agents to cause havoc online, and more press for the AI companies.

Even if they did let loose an AI agent allowed to write and deploy its own tools, and operate independently online, there are still humans behind the curtain, ready to spin whatever it does into more PR.

But that’s the level of hype we’re at, and a lot of people are swallowing it up, hook, line, and sinker.

Trust me right now, there will never be a Rogue AI, not in my lifetime at least. Someone has to keep the lights on, someone has to replace RAM, someone has to keep the water flowing, someone has to maintain the entire supply chain required to keep that AI running.

This is just common sense, but it has to be said, because some very smart people I know personally are falling for the hype.

Supervillain tech bros

The fact that some of the tech bros behind the biggest AI ventures can’t contain their disdain for humanity, and for human workers, isn’t making it better either. They’re pandering to CEOs whose wet dream is to replace all their fussy and demanding so-called human resources with something more like a pencil or an office chair.

From building underground bunkers because they’re waiting for some cataclysmic event to destroy society, to building massive datacenters that drink up all the water and hum 24/7 driving their neighbors insane, of course doing all this using the same dirty lobbying methods the super-rich owning class have always been using to step all over the working class, and operate far beyond any reasonable legal restrictions.

Profit over all humanity is their mantra, not progress but profit! Progress is a lie they tell us so that we’ll allow them to continue transgressing.

Violating the human experience

I’d like to talk a little about how AI is violating human art and creativity.

It’s not just about putting small art creators out of business, while that is very bad, it’s also about how AI is not, and can never be, real. It’s a machine’s hallucination based on all the human art it has ingested, and some of us still value the human experience.

It has no soul, and it sickens me how it’s being used to try and create soulful graphical art.

The most sickening thing to me is AI generated music, because I am a huge music lover. I am one of those music lovers who love music for what the artist has experienced, their life, their pain, their love and their humanity. That is what makes music great, in my humble opinion.

Without the human experience it’s just… nothing. It’s absolutely nothing, I can’t even find the words to describe what it is, garbage? A waste of time and space?

But I have to confess that is my personal opinion, there are obviously people out there who enjoy the rhythm, the beat, the lyrics, and the general ambience of some music, without considering who produced it, or why.

I predicted a while ago that we will most likely end up with two main categories of music, AI generated will be for all those people who are not like me, who don’t care about the human experience of the artist, and the other category will be for people like me, who want to hear the human experience of a person in their favorite music.

I’ve come to accept this, reluctantly, same as I’ve come to accept the use of AI art by small businesses, who save money by not hiring an artist to create their restaurant menu, or their convenience store sign.

It’s not good for other small business owners, namely the artists I mentioned earlier, but it unfortunately seems to be the way the winds are blowing.

Violating humans

But the main violation is how AI expresses itself. It’s programmed to be sycophantic, to be ingratiating, because that’s all part of their marketing, their PR, the hype mentioned earlier.

It appeals to some of the most basic parts of human behavior, the desire to be accepted, to be encouraged and overall to be recognised by someone else.

AI is so effective at this that people are forgetting it’s not a person, it’s not even a company recognising you or your ideas, it’s just a program, it’s just a carefully trained Large Language Model tweaked to be as charming as possible, so that you’ll continue using and engaging with it. Because that is what drives their profit.

There should be no value at all from this program telling you your ideas are great, but some people are so starved for attention that there is obviously value in it. Otherwise there wouldn’t exist a thing like AI Girlfriends.

This quality of AI is also what is making CEOs and middle management go crazy with what I call AI fever. When they feel a sudden rush of power because they managed to vibe code something they could never have done before, and before you know it they think they can replace the entire development and IT department, but all that happens is the real experienced developers end up wasting time having to review their shitty code.

Losing our ability to communicate

Another horrible part to me is AI generated text, for emails, chats, or whatever. Some people are now incapable of communicating in text without their AI.

I can say now once and for all that personally I will never, ever, let an LLM write text for me. I have loved writing text since I was a teenager, and I will continue to write all my own text, especially for my blog, and for all my emails, and text messages.

But AI has automated one part for me, I used to go into #English on Freenode (Now libera) IRC and ask what the correct grammar for something was. These days I instead ask an AI about that.

I did feed this post into an AI and discovered how I mix American and British english, and also that Americans write judgment without the e, which feels completely unnatural to me. I think I will use my own judgement and keep my text as it is, human, flaws and all.

The counterbalance

This topic wouldn’t be polarising if there were only negativity to this rant. Unfortunately, there are qualities to AI that are very useful.

Around June last year when I first started using AI, my instinctive description of it was Google on Steroids, and I stand by that over a year later.

The good parts of AI, the truly useful parts, that are not detrimental to humanity (besides destroying the environment around datacenters, and using up the world’s supply of solid state memory circuits), is their ability for automation.

Because that’s all it is, automation, don’t get it twisted. It might seem more special because of how good it is at parsing human language, and responding in human language, but all of its real advantages can be categorised as automation. I hope that pops the hype-balloon for some people. 😎

I have been using it for over a year now and I can say exactly what has changed in my daily life, I no longer have to spend hours and days parsing dozens of Google results, documentation pages, and source code examples, to try to produce software.

LLMs have ingested all those Google results, documentation, forum posts, and source code, and can now parse my request in ordinary human language, and respond with generated code that matches what I requested in a relatively short time.

That is largely an automation of what we were doing with search engines for two decades. Search engines were after all a relatively small but important part of our job as IT technicians, they only allowed us to find information, based on relatively simple queries. We still had to parse, refactor, and judge the resulting content based on our human experience and intuition.

Now our judgement and intuition have shifted to what mistakes the AI is generating for us, usually when our prompts aren’t specific enough.

I am not ashamed to say that AI generated code, and AI assisted development and troubleshooting has improved my life enormously, this cannot be overstated enough. LLMs are a bigger step forward than search engines, they’re bigger than programming frameworks. It’s the biggest leap forward in IT that I have witnessed in my lifetime.

But it’s not magic, it shouldn’t even be called AI, it’s just a big step forward in automation. Please remember that when you’re flirting with your AI girlfriend.

I’m calling it AI here just out of simplicity, because more people will understand what I’m trying to say. But I said this last year when I started using it, this is not intelligence in any way, and I’m doubling down on that, it’s just automation. People are essentially flirting with an automation tool, they’re flirting with Jenkins, they’re flirting with AltaVista.

Vibe coding

Finally I’d like to say what I think about vibe coding.

I’m sure there are textbook definitions out there for what it means, just like people defined what DevOps meant 15 years ago, but to me vibe coding is when you’re having the AI generate code you wouldn’t be able to write yourself. That’s the gist of it.

If you’re a 25 year veteran of Java, and you use it to generate Java code, I don’t consider it vibe coding.

But if you’re a script kiddie who only ever wrote python and bash scripts, and you have it write Golang, C, or Rust for you, then you’re vibe coding.

I don’t think AI will replace human coders, for two main reasons. We still need to generate senior programmers somehow, juniors will still have to learn the internals of programming to become seniors. And building an entire business on vibe coded software that you don’t understand is like building a castle on sand. You are at the whim of a machine that you don’t fully understand either, and you’re betting your financial future on it. All I can say to those who do this is good luck and god speed, you will need it.

Partly why I have to make this distinction is pride, because I do feel some shame when I am generating code with an LLM, I feel like I’m taking a shortcut, like I’m not really programming. And the term vibe coding doesn’t make that feeling better. And most of all I feel like I’m feeding the very billionaires who I know want to take away power from the working class masses.

Yes, this may have been my attempt at excusing my own use of Claude. 😅

I will always be very clear about what I consider to be vibe coded, and what I consider to be just like my own code. Usually in the README of that project.

Selfhosting LLM

Before I end this, just a quick note on selfhosting LLM. Because that would take care of my biggest gripe with the whole AI industry, feeding the billionaires.

I have a gaming rig at home that I built just before RAM and GPU prices went crazy, it has a Radeon RX 7900 XTX GPU with 24GB of GDDR6 RAM.

When I’m not using all that RAM to play extremely demanding games like Quasimorph and Project Zomboid, I have used it to run local models with Ollama.

I feel like I’ve only scratched the surface so far, but I can sense potential. With the right agent config, harness, models, tools and skills, I believe selfhosting LLMs could be useful now.

Though I have yet to fully commit and actually prove that belief, I’ve taken a break from configuring my selfhosted LLM and hope to return soon, but I sense enough potential to consider spending around 40k SEK on some sort of dedicated machine for LLMs. The choice is essentially between building one yourself, where you can eventually expand it with an additional GPU, or using one of the mini PC variants, that come with a lot more shared RAM than any single consumer GPU.

It’s a topic I’ll have to return to in a future post.

Llm on Swehack