A few years ago, the roadmap felt clear.
Learn programming.
Build projects.
Practice algorithms.
Get hired.
Build a stable career.
That promise b...
For further actions, you may consider blocking this person and/or reporting abuse
"Should you still learn coding?" - of course you should, it's one of the foundational skills as a developer, even when you let AI generate most of the code - how can you judge whether the code AI generated is any good, if you've never written any yourself?
I understand and recognize that people's focus will shift somewhat, but "don't spend time learn coding anymore" (but only learn prompting, and 'architecture', etc) - that doesn't make any sense to me (but, that's also not what you said - you said "coding isn't dead" :-)
Yeah, I agree with you.
Even if AI writes a lot of the code, you still need that foundation to actually understand, evaluate, and trust what it produces; otherwise, you’re just guessing.
The shift feels more like how we code, not whether we should learn it at all.
You can also look at it this way. AI filters out weak engineers from the get go. They think AI will take their jobs. They will be scared off and that means more for us. The engineers that keep on upskilling, adapting, and learning deep will be rewarded. Patience and discipline is what's needed for this generation. Sadly, those two values are scarce nowadays (thanks to social media).
I totally agree with you about upskilling and adapting. That’s definitely what will matter long-term.
I also think deep problem-solving still matters more than ever.
Agree, but how do you acquire problem solving skills if you never start writing any code - by solving Sudoku puzzles, or playing chess? I agree that understanding architecture and so on is important, and will become even more important - but I think that learning how to code (and doing it) is still an important skill, which also helps to develop the other skills ...
Yeah, I agree with that.
As @leob said, you don’t really build problem-solving skills in a vacuum. Coding itself is still one of the best ways to develop them. Even with AI tools, actually writing and struggling through code is what helps you understand how systems work and how to think through problems properly.
So it’s not really coding vs problem-solving, coding is still a core way you develop problem-solving in the first place.
Spot on: "you don’t really build problem-solving skills in a vacuum. Coding itself is still one of the best ways to develop them" ...
I like the memes, gifs and your style of presenting!
Glad you liked it 😄
I try to keep it real (and a bit fun) so it doesn’t feel like another “serious post”.
Appreciate you reading it!
Most of my work is now a) specify what I want, b) AI makes a plan, c) I make modifications to the plan, d) AI implements, e) I make small changes and optimizations and review.
Most of the times it works well. 10% of the times I have to step in and do things myself. Those 10% are the most important
That’s a really accurate way to describe how a lot of workflows are evolving.
The interesting part is exactly what you said; that 10% where things break or need real judgment is often where the most value (and learning) sits. That’s usually where understanding the system still matters more than anything AI can generate.
Precisely.
Imagine letting AI handle everything related to production bugs in payment or banking systems. Hell yeah, then thousands of people could lose money or suddenly have incorrect balances in their bank accounts. 😃
Yeah 😅 that’s exactly where the “just let AI handle it” mindset starts hitting reality very quickly.
In high-stakes systems like payments or banking, understanding what the code is actually doing still matters a lot. One small mistake there can turn into a very expensive problem fast.
The systems of today were built to help people make fewer of the mistakes they used to make when using simpler, less constrained systems. Think C++ vs C, or JS/React vs JS/jQuery. AI doesn't make the same mistakes we do, so it doesn't need all those guardrails. It actually works better without them.
It takes us years to gain the skills to work with these systems; years that won't be available in a world where AI does the coding. Future coders will never have coded a React product from scratch, so they stand little chance of understanding the code AI produces using React in all its glory, or what's wrong when it breaks.
Modern systems are hugely complex and have inconsistencies that require work-arounds, such as needing virtual environments to avoid version incompatibilities. And too often the documentation for a library never got updated when the next version was released. Stack Overflow is full of answers that were once correct but are now completely invalid. We then accuse AI of hallucinating instead of asking why are we asking it to work with these systems when it would produce better results from something far simpler.
Maybe someone will eventually figure out a form of language that makes it both easier for AI to create with and people to validate. Is it too much to ask that we speak the same language?
I took a JS/React proof-of-concept mockup and got Claude Code to translate it into something based on vanilla JS and some home-made tools. No hallucinations, code that's readable by both man and machine and a product that does what was asked. So don't tell me all these advanced frameworks and techniques are needed by AI coders; in most cases they aren't. Worse than that, they actually get in the way of delivering good products.
Yeah that's interesting, I do see your point ...
the vibe coding section hit differently. shipped an AI generated RAG pipeline for a client last year — worked great in testing, fell apart under real query load because the chunking strategy the LLM picked was optimizing for demo scenarios, not production retrieval.
nobody on the team could debug it because nobody understood the retrieval logic. just "it worked when we prompted it."
the 10% Lars mentioned in the comments is exactly where i live now — that's where you either know the system or you're guessing at 2am.
curious if you think the experience gap compounds: does each generation of AI first devs have less capacity to teach the next?
Could't agree more, senior developers didn't come instantly, they were build by deep understanding of trial-error cycle and that is what AI did, bypassing repetitive trial-error