LLMs suck the fun out of coding

Today I stumbled upon this video. Right at the beginning, it mentions “Don’t rewrite, reinvent.” That line could easily summarize what I’m about to write.

I use LLMs daily, and since I retrained into IT during Covid, I kind of see myself as part of the AI generation — in the sense that I adopted AI tools practically in the first months of my new IT career. But I also briefly experienced that creeping feeling of hopelessness when I found the exact same question I was about to post on Stack Overflow — unanswered since 2001. I almost wanted to track the guy down and write: “Hey man, how did it go? Did you ever figure it out? Are you still alive? What should I do? HELP ME!”

Thinking through those first bootcamp assignments, iterating, and slowly working my way toward a solution added a kind of joy you can’t replace. Asking Claude and pasting a snippet just doesn’t feel the same. It solves the problem, sure — but that good feeling is gone. And worse, it slows down learning. At least it did for me, until I started watching out for it. I could write Spring MVC from memory, and I haven’t touched Java or Spring in a while. Why? Because back then, I learned it without AI, trying line after line to understand how to structure a project in a way that made sense.

Anything I later learned with AI only visited my memory — and left without saying goodbye. Yes, it depends on how the learning process is set up. The simple QUESTION → ANSWER → EXAMPLE 🔁 loop never worked for me. I need to take a concrete example, implement it in various situations, and ideally re-create it from memory the next day. And of course, as you probably know yourself, the best way is to come up with your own project and figure things out along the way. These days, I try to use LLMs only when I’ve already failed with trial and error and can’t find the answer in the documentation. That way, things stick better — and I also get used to reading the docs of whatever language I’m learning.

Older languages have the advantage of more resources and maybe wider communities — and sometimes the disadvantage of harder-to-read documentation for less technical people. Newer languages usually have more user-friendly docs with examples and tutorials.

Yeah, it’s less efficient and takes longer, but in the end, I actually understand my own codebase and know where to look — both in my program and in the official documentation.

If you want to quickly hack together an MVP in something you already know, LLMs are awesome. If you’re learning — they can still be great, but be careful.

From my experience, this “AI-first” way of building apps is actually bad for beginners. Writing a good prompt isn’t rocket science, and the LLM spits out a working app. But scaling soon starts breaking down, and in the mess that follows, it’s hard to stay oriented. And when you don’t even know what does what, why something is there, or how it works, it just turns into frustration. Navigating someone else’s codebase is a skill in itself — and it’s not easy. And honestly, there’s no worse feeling than trying to make sense of a foreign codebase that you generated yourself. 😁

My advice: learn slower. Try to figure things out on your own. “Don’t rewrite, reinvent.”

My current flow

1. Watch a short, generic tutorial to get the feel of the tool/language.

2. Open Learn X in Y minutes on the topic.

3. Write a simple program from memory, maybe peeking at step 2 👆

4. Write a few more small programs.

5. Come up with a bigger project and try to build as much as possible using only the docs.

6. At the end, ask the LLM questions like:

   • “Is there a better implementation for this?”
   • “Can this be written in a more low-level way?”
   • “Explain how this works under the hood.”

A program written this way gives me way more joy — and way more skill — than a 30-second AI-generated app.

Negativity to end with

Using LLMs to write LinkedIn posts should be considered a crime, and anyone doing it deserves to be served cappuccino with sour milk at the café.