How do you learn coding today?

I completely understand how you feel. I think many developers are struggling with the same question right now.

My perspective changed when I stopped thinking of AI as "something that writes code for me" and started treating it as an assistant.

You're still the one defining the requirements, making architectural decisions, reviewing the output, debugging issues, and deciding what belongs in your project. AI is helping with implementation, but you're still responsible for the final product.

I also found that I learn much more when I write the first version myself and ask AI to review, improve, or explain specific parts instead of generating everything from scratch. That forces me to understand the code while still benefiting from the speed of AI.

To me, learning programming today isn't about memorizing syntax anymore. It's about understanding how software works well enough to guide AI, recognize incorrect solutions, and make decisions. AI can generate code, but it can't replace your judgment.

AI is a force multiplier, not a substitute for understanding. The better your fundamentals, the more valuable AI becomes.

If you intend to learn coding, don't use LLMs at least while you learn or reduce their use as much as you can.

The reasons:

• LLMs don't provide facts, but aggregates. Look at authoritative trusted sources instead
• LLMs aim to speed up your process of working, which may or may not be preferred when you are already proficient, but learning ought not to be sped up, you need to practice every step of the thinking involved
• LLMs provide you ready-made generated code and transforms you from a coder into a reviewer, so the thing that you learn this way is reviewing, but not writing code
• you need to build your insight on how the code should look-alike and you need to avoid outside influences in order to build this skill

Btw, I do not recommend LLM use even when you learned coding, because you may atrophy your hard-earned skills, hence I do not think LLMs improve coding abilities, its use at coding is to speed up the process, but quality requires effort and time allocation.

One may not always have a choice, but if you want to be a coder, you need to code yourself and, if you enjoy coding, then you will have a natural inclination to code.

Hence, in order to properly learn, read books, watch tutorials and practice a lot. You can use the internet to find information you don't know, but asking for a solution when you already have all the information you need is a no go for the process of learning in my opinion.

If I were learning coding today, I'd avoid the LLM tools, full stop.

When I was an undergraduate, I took the Compiler Design class. The instructor assigned us to code a parser from scratch. "Why?" we asked. "That's what lex and yacc are for!"

But have you ever tried to read and understand the code generated by lex or yacc (or most code generators)? Or worse, try to modify the generated code in a meaningful way? It's practically incomprehensible, and it's long. In about ten seconds, most people give up, and think, "well, I'm confident it works, and that's what's important."

They never understand generated code as well as code they've written themselves.

That's why we coded a parser from scratch. So we could understand how parsers work. How there's a method to designing grammars so they're unambiguous and easier for a compiler to verify. How different classes of parsers can verify more complex languages.

By getting the hands-on experience of struggling with these issues and fixing our code, we understood the principles much better than if we had just used a parser generator and trusted its output.

Using LLM's as a short-cut to learning coding skills is the same, in my opinion. You're actually skipping the learning part, which requires you to be knee-deep in the struggle. You learn more from making mistakes and recovering from them than by having an "assistant" do the work for you.

There have been studies that confirm this. When students use an LLM too early in their studies, they don't retain much of what they supposedly did.

Use an LLM to assist you after you learn the skill so well that doing the task for the hundredth time doesn't teach you anything new. That's when you need automation to improve your productivity.

Thank you for sharing this! Quite insightful! Yes LLM fatigue is real, learning needs to be done without LLMs initially, I agree

YouTube, but not random speaking heads - guys who do something on the screen, and you can repeat. At first, no matter what. Then, you can search for particular things you want to code. On early stages, there is almost always a YT guide for that.

You can also use ChatGPT, but don't make it write the code for you. Write everything yourself, and got there for checks. You can even request hints on your mistakes instead of an instant fix - it will make you try to find a solution yourself for one more cycle before you see the right solution.

If you use Claude (or such) ask it to explain the code snippet and why it picked that specific solution for that specific problem. LLMs (being mindless) don't mind elaborating and providing alternative approaches too.

Any youtuber you'd recommend? Thanks!

Totally makes sense! Thank you!

Yes! I will try doing that! Thank you!

Yep! It is very important to know this early on, or else you find yourself burried with code files you find difficult to navigate through!

Highly recommend using CodeWars (codewars.com). Totally free, with lots and lots of bite-sized exercises organized by difficulty level, and it supports pretty much every modern language. The best part, in my opinion, is that once you've solved a challenge, you can see how other users answered it, which is a great way to learn new and improved ways of doing something.