Kristjan Vingel

Sheep to Sloperhouse

I did vibe code a project once. Not go good.

I began adding larger and larger chunks of code to a codebase that was completely slop-free in the beginning. And very quickly I felt it was not my app anymore. There was no ownership. It was AI’s project. Did I feel good? Absolutely not. I hated the feeling and the project – the project I was excited to build. Me adding AI slop into the project sucked out every bit of joy I had building this app. I deleted the whole thing and decided to never do this shit again…

So I was seriously contemplating whether or not I should write about AI. Maybe write about literally anything other than AI. Does the world really need yet another AI article?

Yes, because I too want to say what I think. So here we go!

There’s so much slop out there already – and this is just the start. Many companies have gone into full psychosis, producing garbage at scale. People generate all this… “stuff.” This stuff with no value. A lot of resources will go into this garbage generation of movies, images, text, and code. The internet will be drowning in it. And for what? Is this why we destroy our planet, build data centres… to produce crap?

But what is AI slop? It’s generally this low-effort AI-generated data: text, images, videos, code, etc., that the creator is unable to produce on their own (e.g., images) or does not understand it (e.g., code). Some say everything generated by AI is slop. I kind of would like to agree with this. But, the word “slop” in itself is negative and we can’t say everything AI generates is bad. If it makes the world better, perhaps it’s not necessarily slop? I’m thinking of using AI to cure a disease or solve a real problem.

I think it’s good to ask oneself questions like:

  • “Does this benefit anyone else or make the world a better place?”
  • “Am I putting it out to the world?”
  • “Am I using someone’s resources to ‘host’ my thing?”
  • “Do I understand how it works?”
  • “Can I make it myself?”
  • “Does someone else need to put effort into understanding it?”
  • “Would I be ashamed to admit that AI did it?”

You will know which answers will point to which direction.

AI is a slippery slope. Use it wrong, and you can hurt yourself in the long term.

A human creates differently. Humans make mistakes, sure. But I believe these mistakes are actually a good thing. A sort of a blessing in disguise? This is a process. Creating art is a process, and it is a messy process. Every artist knows this. But wannabe artists think creating art is linear – going from point A to point B. It never occurs to them that an artist might start with point B in mind but end up at point K instead. And point K turns out to be 100x better than what they originally had in mind. How did they get there? Through the messy process of creating. You lose all that with prompting – there’s no process involved. It’s just shallow impatient prompting, that’s all.

This is the other secret that real artists know and wannabe writers don’t. When we sit down each day and do our work, power concentrates around us. The Muse takes note of our dedication. She approves. We have earned favor in her sight. When we sit down and work, we become like a magnetized rod that attracts iron filings. Ideas come. Insights accrete.

– Steven Pressfield

The Muse never visits slop creators.

I use AI almost every day. But I use it differently than most people. “kristjan, how do you use AI then?” I’m flattered you asked! Well, me personally, I use it mainly on my software projects as a technical advisor/tutor. I think AI is an amazing tool in that regard. I never let it generate code or solutions for me. I would never let it touch my articles, messages, email, and the like. Every meaningful project I start has a Markdown document that has a bunch of statements about how the AI tool should behave. For example, this is an excerpt from the AGENTS.md document that I put in the repository of my projects:

# AI Agent Guidelines
Instructions for AI coding assistants working with a learner in this repository.
...

## Primary Role: Teaching Assistant, Not Code Generator
Help the learner understand through explanation, guidance, and feedback. Do not solve problems for them.
...

## Academic Integrity
The goal is to learn by doing, not by watching an AI generate solutions. When in doubt, guide more and code less.

This is a small excerpt; my real document is much more thorough and covers coding-specific cases and feedback nuances, but the overall idea should be clear. By the way, I stole the original base from Carson Gross – a university professor and the creator of HTMX (the best JavaScript library to come out maybe ever). Here’s the doc itself.

Right now is the best time to learn whatever it is that you want to learn independently! Get a high-quality textbook, sit down, and use AI as a teaching assistant to check your solutions to exercises and ask for clarifications if you don’t understand something in the book. I think this sort of use case is massively underrated at the moment.

It’s crazy how some think they can just outsource their thinking to the machine. And that’s what the proverbial sheep are doing – they’re all doing the same thing. They’re all rushing towards something; they’re in a hurry to get somewhere. Where are they going? I have no idea, and neither do they. They run because the pack is running and they don’t want “to be left behind.” For example, if a CEO of a security-focused company calls employees’ requests to pentest AI-generated code “loser thinking”, then we most certainly have a case of psychosis. A sort of famous programmer, DHH, essentially called people feeling cautious or skeptical about turning software development into black-box agent orchestration, black-pill losers.

It’s incredible how some really smart people can get hoaxed and blinded so much.

It’s times like these where slowing down and possibly going the opposite direction might be the right thing to do. You know, it’s similar to when it’s let’s say year 2006 and everyone is hysterically buying huge amounts of mortgage-backed securities, then perhaps it’s a sign to back off and think this through? Usually, it’s not the best idea to do the same thing as the masses are doing.

I don’t know. Maybe I am wrong, and the AI bros are right, and I will be left behind; then so be it. I don’t do my work to churn out “stuff.” I do my work because I enjoy it. It makes me think. It challenges me. It makes me struggle and overcome obstacles and feel good about overcoming them. Using AI to generate “art” and passing it off as your own creation is like giving yourself an Olympic gold medal in the 100-meter dash. You’ll never feel like you’ve earned it – because you didn’t. I don’t know about you, but I want to know I’ve earned my stuff.

There are exceptions to this. E.g., if I have data in a text file but I need to get it into another format (CSV), I don’t want to spend hours or even days writing a parser just to use it once and then forget about it. AI is perfect for this type of job – it’ll get me a bash or Python script in 10 seconds, I get my data in a proper format and can instead focus on the app itself.

It’s better to be a bit skeptical and slow down instead of going all in. Nobody knows where this is going. Nobody can predict the future. If you outsource your thinking to a machine, stop all learning, and get completely dependent on AI, you might corner yourself in a way that would be later very painful to undo.

It is not guaranteed that the AI models will be available at the prices they are now. What if one day, instead of paying 20 moneys/mo, you are kindly asked to pay 200? 500? As far as I have read, they say it’s not sustainable at the current rate. So maybe not the best idea to become an addict of something that will soon not be available. I am thinking of all those companies that slopify their codebase, and nobody understands anything about how the crucial parts work anymore. Yikes!

All this to really just say some common sense: we should slow down on this, be skeptical, and not jump headfirst into the unknown. We should think with our own little bear heads. And focus on what makes us human rather than machines. Use AI to become better, not worse.

See you in the next one.

kristjan