Over the course of just a few weeks the term “Vibe Coding” has arisen, gained popularity, and received a wave of derision and backlash. The conversation on both sides has captured the current moment in AI and software development perfectly. I didn’t love the “vibe” of Vibe Coding when I tried it (more on that at the end) - but I do love the questions that it’s sparking.
Before I say more, let’s define our word of the day:
Vibe Coding is sitting down with an AI powered editor (Cursor, Windsurf, V0, Lovable, etc.) and prompting your way to working software.
What makes it “Vibe Coding” is that you are really letting the AI do everything. Don’t check the code. When it doesn’t work, no need to figure out why: just tell the AI to fix it. Rinse and repeat your prompting until you get to what you want. No code. Just vibes.
We’ve been looking for this workflow for years.
Predictions that computers would eventually be programming computers have been around almost as long as programming itself. I remember talking to software engineers roughly a decade ago who thought “intent-based programming” - where the developer is describing outcomes and letting a new kind of compiler handle the implementation - was just around the corner. I shared the same thought with an investment analyst a little later and he said “oh, we already see that, we’re calling it ‘no code’.”
The big “no code” wave took the idea outside of the developer world and promised “you can build software without knowing how to code”. The startup world immediately grabbed on to that promise. It feels a lot like the promise of the great westward expansion in the US when homesteaders rushed to claim land. You weren’t guaranteed success - but you had a shot. And the only requirement was determination (well, that and kicking out the native inhabitants, but that’s outside our metaphor here). “No code” offered the same excitement. You don’t need to hire anyone, and you don’t need to know how to code. You can just build something and sell it.
Generative AI poured gasoline on that fire in 2022 when it seemed that - all at once - lots of us were discovering that DaVinci-003 did a decent job of writing SQL. “The next programming language will be English!” headlines declared with confidence. And this is (after a long enough intro - thanks for sticking with me) the first reason that I love vibe coding:
Vibe Coding asks: “Are we there yet?”
Check your LinkedIn feed and you’ll find countless posts from someone who built an app in a day with AI. Then you’ll find just as many people saying they tried and the results were terrible. (You’ll also find plenty of programmers saying “good luck maintaining it” - but we’ll get to that in a minute).
We’ve had models that excel at code generation for a while. We’ve had Github Copilot. We’ve had Devin (whatever happened to Devin?). More recently we got Claude Artifacts and Vercel V0. The “it’s just around the corner” hype for AI driven coding has been growing and growing, but in these last few weeks I’ve seen the most social support ever for “it’s here! We did it! It’s good enough NOW” all under the Vibe Coding banner.
That declaration - and the arguments for and against it - highlight some even bigger questions we should be thinking about in the software world.
Vibe Coding Asks: “Does the code really matter if it works?”
Devs have always argued about this and they always will. We love to talk about which languages or frameworks are good and which ones are garbage. “Cool you built that with ___ but you’re not gonna use it in production are you!?” We argue about how important testing is and what kind of tests to write. We argue about functional vs object-oriented paradigms (remind me which is in vogue today?). Heck we even argue about how to use Git.
But for every tweet/post/article about the best way to build things there’s someone who chimes in “who cares as long as it works” or “I don’t care just ship fast”.
I’d wager many of us sympathise with the latter opinion when we need to get something out the door under a time crunch… and with the former when we look at that code a few months later.
Vibe Coding leans all the way to the side of “I can ship a thing that works - I don’t care if the code is crap.” As a developer who has worked with a lot of awful code (including lots written by my former self) I’m inclined to disagree. My own experience with AI code editors has involved lots of “oh no! Don’t do it like that! Let me rewrite this…”
But I also realize that lots of code my younger, dumber self wrote is still out there in production today. Lots of hugely successful software (including critical infrastructure you use every day) looks like ducktape and popsicle sticks if you peek into the repo. Even in a world where humans write all the code there are mountains of tech debt in every org.
Does it matter if that tech debt comes from AI? Especially if we can ship faster? Especially if we can just brute force fix it with AI like we brute force built it with AI? Swap AI with “cheap offshore dev” or “quick and dirty startup code” and this isn’t a new question… but the way we answer it is going to have a big impact on AI adoption.
Vibe Coding Asks: “Is Code still Craft?”
If we compare building software to building a house - is the programmer the construction worker or the architect? If you’re a programmer you probably just said “both! And the structural engineer! And the…”
Programming is not just “writing code”, it’s deciding what code to write and how. What paradigms you’ll follow so that it all makes sense together. You’re considering where it will need to run, under what constraints, and a million other things. And there’s more than one right way to do it. Coding is art - it’s a craft as much as it is a science.
Vibe Coding claims to separate the art from the science. This is one of the more common takes I hear from experienced developers using Cursor or Windsurf. It goes something like “You shouldn’t let AI write code you couldn’t write yourself. But if you know enough to tell it how you want things done, it really speeds you up.” The thinking is that you can continue to play the architect while you let AI do the “manual labor”.
Like the last question, this is one we’ve wrestled with long before AI was writing code. It’s part of how we reason around hiring, the development of new hires, the role of staff engineers, etc. But it’s coding agents that have got us talking loud about it.
Vibe Coding Asks: “Who controls the means of Production?”
Now we get into the interesting economics of opportunity.
Let’s buy all the hype for a moment and assume that yes - we are there - you can Vibe Code your way to a unicorn.
Who is this good for?
Well, it’s good for anyone who’s ever said “I’ve got a great idea - I just need someone to build it!” Great. That someone is AI now. And with every cloud provider throwing free credits at you, the path to launching something is pretty much free (don’t worry if you don’t know how to configure a server, AI can tell you how to use all the popular turnkey services.) I have to admit there is something really exciting and democratizing about that.
At the same time, my career has been built on being the “someone to build it.” For now I can tell myself “don’t worry, when they need to scale they’ll need me”. But if I accept that you can go from 0 - 1 with AI now… how long until you can go from 1 - N?
So all this exciting opportunity that comes with “anyone can build something”... does it mean loss of opportunity to generations that have been told “learn to code - that’s the ticket?” Do we become a dime a dozen?
This makes me think of AI image generators. It’s easy to see why artists would be up in arms against a threat to their livelihood (regardless of who had the right to train a model on those images) - but I can’t deny the thrill I got the first time I put up a homepage with great “illustrations” on it that I didn’t have a budget for.
Every advancement of AI seems to bring this double-edged sword. It promises to unlock things like art, code, even legal advice and give them out to the “masses” - people who just couldn’t afford to hire a specialist. But it threatens the income streams of the humans that provide those services.
And who will profit most from that democratization? Will we all rely on a few huge model providers to deliver it - shifting more and more human labor to their compute? Or will open source keep that from total centralization?
Vibe Coding Asks: “What becomes of programmers?”
Would you advise a high school senior today to go to school for software engineering?
For most of my career I’d answer with a resounding “yes!”
In the last few years I’ve been less and less sure, not only because of AI but simply because of how many of us there are. An attractive line of work has - not surprisingly - attracted a lot of workers over time. I’m amazed at how often I find incredibly talented developers willing to work for far less comp and far more hours than I am. I wonder “could I compete if I were starting out today?”
Vibe Coding may be a term already on its way out by the time you read this article. But the conversations around it are circling the key question:
“With or without AI… has ‘writing code’ been commoditized?”
I am forever the optimist. I believe that while paradigms may change and abstractions become more abstract - sharp, intelligent, skilled human beings will always find opportunity.
But I’m forever self-doubting too. I can’t see the future any better than you can. So I’m listening closely right now.
The Author Asks: “What about fun?”
Don’t knock it ‘til you’ve tried it right?
I had to give Vibe Coding a whirl myself before sharing my thoughts. And yes, I was able to build an app in a day. Some of the code was great. Some of it was a mess. But it worked!
Often I was frustrated by Windsurf’s agent rewriting a module that was completely out of scope, or misunderstanding the cause of a bug and doggedly trying to fix something that was already working over and over again. But it still got done more in that time than I could have “manually.”
Would that app hold up in the wild? What about when I want to add features? How many vulnerabilities are hidden in that code? Who knows. (And that’s part of the context here - vibe built apps are a bit like “50 year shingles” that have only been on the market for 5 years…)
But my biggest takeaway is that it just didn’t feel good.
Oh it was a blast at first. I just told the AI what I wanted and got something working so fast. That was a thrill!
But there is a feeling you get when you spend the whole day programming. You hit your groove, or flow state, or whatever you want to call it. It’s a good day. You get a lot done. And you leave your desk feeling like time flew by and you just built Rome in a day. There’s a smile you can’t wipe off your face.
After about half a workday of “Vibe Coding” the end result was cool, but my overall feeling was more like the one you get from a day full of meetings - one where you spend the half hour breaks you get in between scrolling LinkedIn because there’s “not enough time to start anything.” I was tired, distracted, and didn’t really feel like I’d accomplished much at all.
Maybe that’s just my brain.
I got into programming because I like it after all.
But for me, what Vibe Coding was missing… was the vibe.
