Has vibe coding culture done more good than bad for the industry?
Posted by Look_for_some_stuff@reddit | ExperiencedDevs | View on Reddit | 46 comments
Vibe coding and AI in general really lowered the barrier to starting coding. People do ship real tools to solve real problems. But too often, it's just some non-tech people who create code that they don't actually know and can't debug. And often the code is riddled with vulnerabilities that can put the data at risk.
It can still be a good starting point for people who are only beginning to code. But somehow, I feel it will be only a fraction of current vibe coders. The rest will just keep shipping things they can't maintain or secure
exomyth@reddit
Let me change this old joke:
You'll have 10 types of vibe coders, those we do and those who don't understand this joke.
Those who do, generally just supercharge their ability to learn and build what they want, those who don't will plateau, and be less successful then those who do
PatchSprite@reddit
The barrier to entry dropping is genuinely good. More people solving their own problems with code is a net positive.
The issue isn't vibe coding itself, it's that shipping has become decoupled from understanding. A lower barrier to entry is fine. A lower barrier to production is where it gets dangerous.
The vulnerabilities problem especially. Someone who doesn't know what SQL injection is can now ship a user-facing app with a database in an afternoon. That's not a hypothetical risk.
The optimistic take is that this creates more demand for engineers who actually understand what's happening under the hood, someone has to clean this up eventually.
EntropyRX@reddit
It’s been a net negative. The illusion of productivity. The illusion of velocity. POCs and MVPs disguised as products. AI slope everywhere. Everything pushed with AI requires endless cycles of rebuilding due to edge cases breaking the whole system, endless alignment meetings as no one really understands what’s going on.
Mark my words, we’ll soon start seeing corporations claiming NOT to use LLMs to develop their software to both attract talent and reassure their clients.
nimzobogo@reddit
Nah because each release, the things get better and better.
EntropyRX@reddit
Sure, and agi is 5 months away.
nimzobogo@reddit
No, LLMs won't achieve AGI. But that's separate from whether or not they can code better than most SWEs. Each release, they get more capable.
BigDickedAngel@reddit
Its a bad starting point to learn how to code...most of the learning an engineer does is from being wrong and having to figure out why.
ninetofivedev@reddit
Is there a subreddit for experienced progressionals that isn’t full of AI doomers?
chat-lu@reddit
Aren't the doomers the ones who claim that AGI is near, will become Skynet and doom humanity?
We think that LLMs are dumb and will rot your brain. It's a very different perspective.
ninetofivedev@reddit
Different doomer
chat-lu@reddit
Where’s the doom? Nature will heal after the bubble bursts.
ninetofivedev@reddit
This isn’t blockchain. This is cloud. It’s not going away.
chat-lu@reddit
Who will pay for it?
ninetofivedev@reddit
Who pays for your cloud bill?
chat-lu@reddit
My employer because it finds it’s worth the price. LLMs are heavily subsidized. No one finds the tokens worth the price.
So who will pay for them if they are not going away?
ninetofivedev@reddit
You think cloud wasn’t heavily subsidized in the early days? Remember when cloud providers used to pay you massive sums of money to switch to their cloud?
chat-lu@reddit
No it wasn’t. People saw the value rather quickly. Nothing has ever been subsidized to that point.
ninetofivedev@reddit
You’re so close dude.
chat-lu@reddit
You’re not making any sense. If you don’t say what you think will change so that businesses that aren’t willing to pay anywhere close to the full price suddenly do, I’m leaving this silly conversation.
ninetofivedev@reddit
I’m not making any sense because your cognitive dissonance won’t let you understand that the same is true for ai.
chat-lu@reddit
Bye
Gloomy_Cicada1424@reddit
honestly tools like GPT, Cursor, Runable etc have probably gotten way more people into building things than ever before which i think is genuinely a good thing
the downside is a lot of ppl can now generate working code faster than they can actually understand/debug it 😭
feels like the real bottleneck shifted from “can u build it” to “can u maintain and reason about it 3 months later”
bleckToTheMax@reddit
I think my biggest struggle with it is that's it's been coupled with job insecurity. If I felt as economically secure as I did a few years ago, I think I'd be much more happy with AI usage.
nimzobogo@reddit
Exactly
hyrumwhite@reddit
It hasn’t lowered the barrier at all. It’s lowered the perception of a barrier
nimzobogo@reddit
It has for sure raised the floor. Kind of like the difference between a chef and a cook, though. Anyone can be a cook. Being a chef requires a lot of training and know how.
throwaway_0x90@reddit
You know this is just going to attract unproductive arguing.
chat-lu@reddit
With whose money?
originalchronoguy@reddit
This isn't a zero-sum game. There will be winners and losers. Both have valid argument. For the naysayers, I agree with you. I have to compete with non-technical people vibe coding Base44. And to combat that, we do agentic orchestration . Is it vibe coded? We can argue that all day but you have to fight fire with fire.
That is how you survive. You have to fight with lowest common denominator. Back in 2010s, people were offshoring to $3/hour engineers overseas. We had to fight and combat against that. We won. So it is the same battle, different opponents.
When a vibe coded app was done in 2 weeks by a VP. I will produce a secure, governance aligned version in 3 months. Versus the luddites who will build it in 8 months the old way. This is how you win battles. Leadership, at where I work, are reasonable enough to know that 2 week base44 app is not prod ready.
The biggest win, whether you hate AI or not is I can do POC that is better than any Figma or mockup. The POC works with all the data flow logic. Even if it is writing to a SQL lite db for the time being. It proceses non-queue items. I can push a button and get a widget out. You can't do that with Figma. So the POC becomes the user requirement -- match parity with it. You click 13 different buttons in a unique data flow, you get the same deterministic output -- a video, an excel file, a PDF. whatever. But the real prod product needs to scale to 10,000 of users. This is where the win is. POC gives technical validation so real prod engineering can happen. That is an accelerant even if you don't use AI generated code.
So I have no problem with a VP vibe coding a demo, If the demo is functional, I will re-write it from the ground up very quickly. Those are my user stories. Make that demo a working product. No different from a data scientist team doing demos in Jupyter notebook with .CSV and converting that to a Flask API to do real time prod ingestion with a data lake. They just call that process MLOps. Maybe we can have LLMOps now.
ninetofivedev@reddit
The offshoring battle never ended and it started long before 2013.
AI puts out exponentially better code than Indian contractor farms. And I don’t have to wait 24hours to find out the misinterpreted my requirements.
w-lfpup@reddit
Vibe coding is bad mkay
roger_ducky@reddit
There’s no culture yet… I mean, most anyone can do this is about 3 years at most.
Most non-technical people can only do it for the last year.
Always good to have people that know the business requirements actually specify something concrete.
The trick is to do it in such a way that SW people can actually use it as a module for the production ready version.
bruno_pinto90@reddit
Today, i got assigned a ticket of an error thrown only one sample of 12 and only 3 times in a matter of 3 weeks. Good luck Claude!
boring_pants@reddit
Is this a serious question?
Antique-Stand-4920@reddit
I worked with a more tech-savvy client who used vibe coding to help explain his requirements. It was actually very helpful. That said, our dev team only used it as a reference and implemented the real product via good engineering practices. I think if vibe coding is used that way, it could at least help the tedious process of getting concrete requirements.
thefox400@reddit
I agree that it has lowered the barrier to entry, which should be a good thing, but it also lowers the ongoing quality for everyone using it. It also seems to discourage growth. Lowering the barrier to entry is only a good thing if, once entering, individuals have the same growth as they would have before. My 2 cents are that vibe coding is "training" coders like AI art is "training" painters... it's not.
Miserable_Heron_9007@reddit
So far, milk, bread, and gas haven't got cheaper (the other way, in fact; maybe if there wasn't the wars things were different).
nomoreplsthx@reddit
Vibe coding has done massive harm, but that's sort of a technicality. I wouldn't call responsible AI use 'vibe coding'.
It's like the problem with saying murder is wrong. If killing isn't wrong it isn't murder. It's self defense or whatever other form of justified killing. So it's tautological.
Similarly, if the AI use is resonsible it isn't vibe coding. It's AI prototyping, or AI assisted development, or using AI to build a toy. Like, it's obvious that someone with no ability to understand what they built deploying important production code is bad.
ninetofivedev@reddit
Vibe coding has come to be the term for any usage of an AI coding agent.
At the surface, it's a meaningless term.
Groove-Theory@reddit
Vibe coding has done something very very dangerous: it made overconfident non-technical dickheads even MORE overconfident. More overconfident in technology that they don't understand. They can't tell you anything about how a neural network works in principle, anything about transformer architecture, anything about LLM attention....nothing. They think vibe coding is just some magic black box that they can just shit out all their unrefined dumb fucking business rules. And the model says "you got it boss!".
It's awful. There's a reason why engineers have historicallly been dickheads about getting things right and being annoying about it. It was one of our few safety nets to keeping our work somewhat sane.
Ok_Body7659@reddit
Vide Coding? That's a bad bad bad word in my book. What I've seen it do, is give allot of power to people who don't understand the underlying technologies being used. I don't see this as a good thing as it takes some discipline and good skills to be fast and understand all the code. And the security vulnerabilities is a huge one.
On the other hand, AI in general, in the hands of a good developer, can be very powerful. But the question is, how do you separate the good devs from the vibe coding devs. Token usage? I feel this is a little more difficult for those that don't know code.
I agree, a good starting point, or for developing a proof of concept. Problem is, sometimes those POC's go straight to production.
Quarksperre@reddit
I think it is in the process of wasting a whole generation of programmers.
Good for me. Not so good for all those guys.
Its now about four to five years in and I don't expect true AGI emerge in any form anytime soon. And especially not with LLM's
I have troubles myself to not actually atrophy my skills because some things became so easy. But I have some "skill points" to spend.
Not so much with Juniors. They don't have skill points and they never build it.
So either AGI and all those issues are magically solved and we are obsolete. Or its just a bunch of guys without any valuable skill points.
DanManPanther@reddit
If we split vibe coding and software engineering with LLMs, I think vibe coding is doing more harm than good.
There's 2 ways you might split this:
It goes from Vibe Coding to SWE with LLMs when you understand what the LLM is doing and producing.
It becomes SWE with LLMs when you put the right level of effort, detail, knowledge, and skill into producing specs and methods for the LLM to self-validate.
There's a pretty big difference in a designer building a stateful web app with Claude, and a senior software engineer with 10+ years of experience using Claude to build a stateful web app faster than they would otherwise with Claude, able to manually change code, check for flaws and risks, etc. Someone who can pick up and fix or finish what the LLM is unable to do well enough to ship something secure and scalable.
One big problem with vibe coding is it makes non engineers think less of engineers, and push to cut corners or take large risks without being mindful of the tradeoffs. (It also leads to just, awful products that launch broken and insecure).
I don't see a problem with going faster with AI as long as the hands at the tiller are experienced and capable. We also need to be realistic about the speed up. We don't really understand how to effectively estimate this kind of work yet (as an industry). It also seems to vary considerably by the type of task. Some problems can now be solved in a fraction of the time, some the same, and some will be slowed down by the use of LLMs.
The other big problem is skills atrophying or never being built. We have fewer room for junior engineers and we force them to "go faster" with AI without taking the time to deepen their skills or knowledge. That is really going to bite us when there's suddenly far fewer seniors, and they are on the whole less capable. We need to make room to address knowledge and skill debt.
If_I_Could_Just@reddit
Just feels like a dam broke and what’s easy to do is represented by where the water immediately rushes. Value is in where it can’t easily go. Vibe generated code is the easy flow - not much value. Engineering with value shifts to where things aren’t so easy.
mehtheswede@reddit
I have been doing this for almost 20 years and have never had more fun at work or at home than in the past 6 months.
I’d argue it is a net good
SplendidPunkinButter@reddit
Absolutely not