Table of Contents
Introduction
AI Is Awesome
Failing Code and a Cold Bus
And Here Comes the "However"
The Name Itself
The A...
For further actions, you may consider blocking this person and/or reporting abuse
Wow such an explanation, loved each phrase and the way you described our old way of solving problems. Also agree, that it's easy to prompt and build a project with AI but maintaining is a skill which comes to developers through experience and constant learning.
Appreciate that! Thanks! 🙏
Here are my thoughts on vibe coding after giving it a serious try building a static website for a week (Claude Sonnet 4 agent mode via GitHub Copilot):
Final grade: D+
IMHO, vibe coding in its current state has its uses in proof of concept or toy projects, but it's not ready for prime time. Since this was my first experience with vibe coding, I kept a mental journal. These are the various emotional states I went through:
I feel that if I were to vibe code as a full-time job, I would experience brain rot, as the AI is doing most of the thinking. You are on the sidelines, waiting for it to respond, and then trying your best to steer it in the right direction. You are not learning anything new in terms of programming. You might at first, but that will eventually give way to a routine where you just want to try out the generated code without fully understanding it.
Furthermore, I feel that vibe coding might even reduce my lifespan because I found myself growing increasingly exasperated with it as it frequently kept getting in its own way. The stress factor is real, and it's not the vibe that I want.
This comment can make a full article. Thank you for such detailed insights. I experienced more or less the same effects when trying out vibe coding — I was stressed out at the end of the process, which was a different kind of stress than when working on your own code.
Thanks, that was the CliffsNotes version. 😅 Maybe I'm old school (started with Turbo Pascal), but I think I'll continue to sculpt my applications line by line and fiddle with new technologies myself. It's more fun, less stressful, and you learn more. With that said, I'll continue to dip my feet in the AI waters from time to time because I'm optimistic that it will improve faster than Moore's Law. Exciting times.
"vibe coding" is a term used to describe letting AI go wild to build weekend toy projects. That's the pure definition of it. It's different from "Tab Coding" where the code is mostly the developer's work.
I disagree, you're doing something wrong on your end, while I experienced what you describe, I don't experience anything you mentioned and at $100/m a month it's dirt cheap compared to having a junior dev. I've created 4 full stack web apps in the last month and a half. As long as you stick to NORMAL Dev standards you're gonna be alright.
One of the points of the article is that vibe coding is very easy if the problems are very easy. What do your full stack applications do beyond basic crud and it's supporting UI (e.g. forms to input the data, buttons to save or delete/cancel)?
Well if it works for you, good for you. That’s great!
I'm glad it's working for you 👍. While I may sound like a non-believer, I'm optimistic that someday we'll all be able to lie in bed staring at the ceiling instructing our AI agents how to perfect the app that we dreamt up yesterday.
Whilst a vibe coding can produce something quick. You can also use templates to do the same. I used AstroJS templates and they crested wonderful websites quickly, but it was well engineered and designed.
🙌 Thank you! I couldn't agree more about the whole vibe-coding thing (and I'm just as much a fan of the name). You're right though - it DOES have it's usefulness and there's a time and a place to let AI run wild (production isn't one of them).
This is my version of, let's call it "vibe-adjacent" coding (because it IS really fun):
So why do it in the first place?
Had my goal been to build a prod-ready, enterprise scalable, fully secure app, then I would have just written one. What I wanted was to understand how to utilize AI, learn which prompts return the desired results, and have something shareable at the end. No - I didn't write the code. I spent weeks instead writing prompts and instructions, testing different AI models, slowly over time giving it more and more freedom just to see what happens when you push it to it's limits in that kind of environment.
That's about as much "vibe" that goes into any one of my projects!
The use cases in your examples are perfect. I think as long as you're learning along the way and helping others when you can, then have at it. The whole "I'm not a developer, but I built <insert thing of the week> with AI" thing? Drives me insane, too! (bonus if they throw in a "their job isn't THAT hard" or "I don't understand why they make xyz" somewhere)
I'm glad I'm not the only one 🤝
Thanks for sharing such thoughtful insight. Your comment alone (with a few additions) could easily make an article on its own.
I completely agree that vibe coding can end up being more time-consuming than just development, especially when you consider long-term maintenance.
I’m really glad to hear the post was relatable to you! 🙏
I'll share something interesting that I found on X. It's called Omakase Software.
While I can vibe-code and make something useful, that doesn't means that I should.
Also your post is on point and vibes really well. 🔥
That's a very interesting take. Thanks for sharing!
Love how you capture the importance of the 'journey' in coding - it really is where most of the growth happens for me, too.
I'm finding that when AI tools are built to keep context and let you gradually shape things (vs just spitting out code), they can actually deepen that journey instead of replacing it—curious if you think we'll get more of those tools soon?
Thank you! The AI tools are definitely here to stay and I think more of them will be added in the future — they definitely help us in many ways. I use them almost every day.
I can even state that AI tools made me a better developer in some ways, however we need to watch out not to overuse them, just like anything else in the world.
I’ve just learned 2 new things about me that I didn’t know before:
Thank you!😄
No he meant Dotallio not you! lol! Amazing article xx
The way I use AI is not unique, but often overlooked. I'm an individual that can get overwhelmed by the shear amount there is to do on a particular feature, module, or project. That coupled with an unhealthy serving of perfectionism can keep me from getting started on things for longer than necessary. AI, specifically some copilot-like code completion circumvents those issues because "getting started" is as easy as opening my editor and placing my cursor, and I don't get caught up on perfect because I expect the AI to generate slop. Often times I'll wind up completely ignoring the AI's suggestion because I thought of a better approach, but that approach did not occur to me until I saw a way I knew was insufficient. It also lowers the mental barrier for reworking code since it was not a long arduous fight with myself to write it in the first place. I'm less attached to the code in a way that makes me hesitate to rip it out and redo it better.
That's a good point you're making. Every tool needs to be used in a smart way to get the most out of it. AI is no exception.
Pretty much this article describes what I was saying for literally months: AI might make you lazy. Yes, it generated code accordingly to your prompts but eventually you will hit a technical wall when AI starts to pollute the expected results with something that is not planned. It happens A LOT and if left unchecked, it will be converted into technical debt that will be hard to assess. How hard you might ask, it depends on how many layers or abstraction AI wrote to solve a specific problem.
Sometimes, the solution is ridiculously one when you could simply overwrite hundreds of abstractions with few ones once a engineer sees the problems. AI won't replace a human engineer in the long term because of this. It lacks vision and clear understanding of the general goal. AI is excellent for short simple tasks, but not for projects.
One more thing. When building short tasks with AI is where you will find a great reward by learning from those small solutions.
Exactly! I like how you summed it up.
Great article! Summed vibe coding up so well. I'm so grateful that I learnt to code in 2021 just before LLMs were available as a junior developer still in the world of Stack overflow 😅 I agree the stress is worth the end result! "Nothing that is good comes easy and nothing easy is good". I could definitely see if I was a junior developer today that vibe coding would be so easy to use as a saviour and not fully be aware of the risks.
Oh, so you just barely missed the "I can generate my entire enterprise app without writing a single line of code" era. congrats! 😄
To be honest, I worry about the new wave of junior developers entering the field now. Sure, information is way more accessible than ever thanks to LLMs, and that’s definitely a good thing if used wisely. But when you’re just starting out at your first company, you naturally want to prove your point by being fast, sharp, and seemingly capable of everything. So I can’t really blame them for overusing AI tools and skipping the deep dives where real understanding is built. I remember how desperate I was to solve complex problems when I was a beginner and pretty much everything was complex for me — the stress was unbearable. But still, let’s see how it all unfolds.
And you made a fantastic point that “nothing that is good comes easy, and nothing easy is good.” There's no replacement for hard work, no matter what kind of tools you have at your disposal. I always remind people (including myself) that the pain and struggle will eventually pass but the success will stay with you.
Great read! Your 1980s job listing example had me cracking up, that really drives home how absurd the whole thing is when you strip away the tech hype. The part about hitting walls with complex enterprise stuff really resonates too, I've seen so many projects that started with AI magic and ended with developers pulling their hair out trying to untangle the mess.
Glad you liked it! That 80's example is my favorite part. 😄
this is extremely impressive, the amount of stories and detail here lines up so much with my own early coding days
you think there’ll ever be a real way to blend deep learning and hands-on work so you don’t lose the vibe
Thank you! I’ve been outlining and refining this article for over a month — I didn’t want it to feel shallow. I wanted it to be relatable. I’m glad I was able to get my point across.
I truly believe there has to be some ways for balancing between deep learning and hands-on work. We’re at our most effective when we get the most out of both. And software engineering is all about effectiveness.
So well written! I went through the process of joining just to be able to comment on this article!!
I've made a few sites using "vibe coding"
JacketOracle.com
MergeNerd.com
UnicornBookClub.com
Oh and a . XML transformation file which was so fast and works great. Anyways I just want to say that it's really opened up the doors for me to peer behind the curtain. In the last couple months I've seen these projects from idea to deployment (none are popular -which is ok by me) and digging in I've learned so much.
For simple websites and tools AI can really shine. Keep up doing good work!
I came across this blog through the Top 7 Featured DEV Posts of the Week — and I’m glad I did!
It’s such an easy and engaging read. Even for someone not deeply into software engineering, the way you explained things makes it super approachable and interesting. I read it all the way through, which doesn’t always happen with technical blogs
Thanks! Tried my best to make it as engaging as possible. It was harder than I thought, but totally worth it!
Yeah its a stupid name and a joke if you live in Europe because flight provider Jet2 have 'Vibe' package holidays which is always adertised everywhere. Someone should update wikipedia and rename it to 'Generative Development' or 'Gen Dev' for short.
Haven't though about that to be honest 😆The name you've suggested could work. At least it's way more self-descriptive.
I've been programming since 1998, I started as a kid, 7 years old.
The "VIBE" coding trend(not AI in and of itself) is scary and depressing. I pretty much agree with all your arguments.
Since 1998, as a 7 years old? That’s impressive! Such an experience! Which languages have you used and which are your favorites?
I'm that guy. 65 never coded in my life. I took a COBAL class one time 25 years ago. Now I find myself addicted to vibe coding and trying to build out an entire product system in my spare time. I don't know who said it recently but I read that ""AI won't take over humanity, it will just make it so easy for humanity that humanity will just hand it over it AI." I believe it is already begun & humanity is the frog in the the heating pot.
That is awesome! It’s never too late to try out new things and you’re a perfect example of it. Who knows, vibe coding might even encourage you to start learning coding again—that would be amazing.
Good luck and keep vibe coding!
I've tried Vibe coding. Yeah it worked, sorta. I made a checkers game in Python. Even after spending over 6 hours prompting, it still has a few bugs. It's fun as a throwaway kind of project where your life and career don't depend on it. But here's the thing, I tried looking into solving the bugs, but I'm not advanced enough yet in Python, plus if I have to start spending time understanding the code, working my way debugging through each function, trying to understand intent, etc...what have I actually accomplished?
There will be an AI reckoning in another year or two. The bloom is off the rose. Everybody that thinks they can just one-shot their software and forget about SaaS, forget about software design principles, interoperability, security, etc... they're just going to make another Y2K scenario. A bunch of startups are gonna flop when the VC money dries up, when the VCs realize that having 5 million custom made copies of "almost the same" software probably isn't that good for the ecosystem. It's just a matter of time
That's spot on! Totally aligns with my thoughts.
I think your definition of “vibe coding” is spot on but I’d like to share a small counter-example from my own experience.
I’m developing Pagonic, an alternative to WinRAR. Since day one, I’ve been using AI as a support tool but I’ve never fallen into the “AI writes, I just watch” trap. Quite the opposite: I manage AI with strict rules and structured plans. My sprint documents even include a dedicated AI input window that clearly defines which decisions are delegated to AI and which ones I review manually.
For instance, I used AI to help kickstart my parallel extraction engine, but every step was supported by thorough testing, benchmarking, and postmortem analysis. At this point, AI has helped me move faster but the control has always remained in my hands.
In my view, AI isn’t the problem the issue is using it without boundaries or structure. Your post really nails that distinction. Thanks for sharing it
Totally true! If the steering wheel is still in your hands, you're good — you can use AI as much as you want, it'll only make you more productive. When AI takes over control, I don't think this is the spot anyone wants to be at — well at least for now.
It's not "Vibe Coding" if AI generated code alone is used in production. "Vibe Coding" is a term used to describe letting AI do it's thing with minimum human work to develop weekend toy projects.
Thanks Giorgi, it was totally awesome to read such a balanced, insightful piece of writing, you made my day 😀
Thank you!
George, you nailed it! I'm not "All In on AI" at all. I'm still quite skeptical about the current LLM-based AI assistant hype. Still, I'm using it, sometimes struggling, sometimes taking important steps forward to better code and new content creation ideas.
I had already collected some ideas for a new blog post about recent AI coding experiments, "real vibe coding" === "getting into the zone", and what has happened to Music Monday, but I had turned away a bit from DEV after its announcement to collaborate with Google AI. Posts like yours keep me coming back to DEV and inspire me to finish and publish my post drafts!
Where is the Vibe in Vibe Coding and what happened to Music Monday?
Ingo Steinke, web developer ・ Jun 25
Thanks for sharing! I’ll definitely check out your article.🙌
Putting environmental sustainability to one side... I'm liking the look of Harness. Vibe testing, if you will...
Created an account just to appreciate your writing! Spot on, my rancid friend.
This is a straw man. I don't think a lot of people literally do this... it's not usually even an option. If the way you choose to define vibe coding is as literally not even looking at the code then what's the point of even writing this article? It's like writing an article about the dangers of living in a large city but your focus was the tiny fraction of people who do parkour off of buildings.
Hmm, I've been using AI coding for 4 months now and while I experienced what you are talking as a beginner you learn to circumvent almost all the problems you are talking about. You have things called knowledge files, use them.
I wonder if this is what people who coded in assembly said about compiled languages. That was before my time. But I was around to hear people who used compiled languages like C or Fortran make similar arguments about interpreted languages like Python.
Growth like this is always nice to see. Kinda makes me wonder - what keeps stuff going long-term? Like, beyond just the early hype?
That’s a great question. I think the quick answer would be results. Time is the best tester.
I think it is a mistake that you're grouping PHP into a legacy language like COBOL.
PHP is not a legacy language but PHP 7/8 is a modern language that in my opinion is a better choice to Java and C#. I write an enterprise system on PHP, and it's fast, reliable, and maintainable.
I don't think vibe coding is as big as what it's made out to be. Not many have the desire or persistence to hammer on your ai to get something to work, especially when you don't know your subject...
deep researched done by AI. Isn't it??
The section about average energy consumption is actually from ChatGPT, since it was specifically about ChatGPT itself — I didn’t really know the exact numbers.
The rest is mostly based on my own research and thoughts. It took quite a while to gather everything and shape it into something representative but it was definitely worth the effort.
Google’s emissions up over 50% due to AI in data centres. It's resource consumption is huge and it's not making a profit. share.google/vHNETVrjqtjvZCj3G