The Illusion of Vibe Coding: There Are No Shortcuts to Mastery
Posted by shift_devs@reddit | programming | View on Reddit | 154 comments
Posted by shift_devs@reddit | programming | View on Reddit | 154 comments
rashnull@reddit
They’re not Coders. They are Vibers!
BlueGoliath@reddit
Who vibe codes and cares about mastering coding?
mouse_8b@reddit
I already mastered coding. That's why I vibe code past all the boring stuff.
Kinglink@reddit
Serious question, do you read and consider the output of the AI? Or just check it in/run it with out testing it?
mouse_8b@reddit
Yeah, I read it. I don't turn in any code without running it and testing it. I didn't get to senior level by turning in crap.
mfizzled@reddit
is vibe coding not frowned upon as a senior? not saying it can't be used as a powerful tool for an experienced dev, but it does seem like it would raise a few eyebrows in places I've worked
cplusequals@reddit
As a lead, spending half a day writing CRUD endpoints and tests is frowned upon because it's a waste of time. Let the computer do that. Focus on making sure the business logic is correct and does what you need it to do.
Scatoogle@reddit
Who is spending half a day writing end points?
cplusequals@reddit
The better designed an architecture is the more code is boiler plate. Seems like a universal experience for seasoned developers to have been swamped under similar level code fairly frequently, no?
Scatoogle@reddit
If you are following half decent modern design practices your endpoints should be the simplest and fastest part of your code. You also shouldn't be writing more than like 2 or 3 per task. If you are spending half a day writing code you are either profoundly slow or doing too much work at once. And as a senior level engineer you should be pushing these simpler level tasks to your newest or most junior level developers so they build knowledge of the code base.
cplusequals@reddit
I'm not going to have any of my devs writing boilerplate code because it's a massive waste of time and money. They're all on business logic. AI handles the grunt work. Which, by the way, you shouldn't pawn off on new employees or you'll stunt their growth if you're in the unfortunate position where you're not allowed to use AI or other utilities like a modern IDE.
Scatoogle@reddit
You give them those tasks because it puts them in a position to become familiar with a code base. There is more to just writing method stubs that goes into an endpoint implementation. They need to call the write services, set endpoint permissions, depending on the size of your team they might need to do a basic bitch hookup with the front end of testing. Just generating an endpoint alone takes all of like 10 seconds. And using AI to generate stubs is extremely hit or miss. I can't get Rider with Claude to generate CPP header files or C# interfaces that are actually what I asked for.
cplusequals@reddit
That makes more sense for a new hire than a lower titled dev which is probably already familiar with the codebase.
Wouldn't know. I've never had success with Copilot either. We use in house models devops manages. They work very well and save an enormous amount of time.
mfizzled@reddit
true, although I wouldn't expect a lead to be writing stuff like that when a junior/mid could
mouse_8b@reddit
I guess your company just has a bunch of jr devs hanging around waiting for work?
mfizzled@reddit
Why would you guess that?
mouse_8b@reddit
Just seemed like something someone with a bunch of jr devs would say
cplusequals@reddit
Sorry, let me clarify. I meant that, as a lead, I don't want to hear that my juniors/mids are wasting time on tasks that have already been automated.
Kinglink@reddit
It should be heavily frowned upon. I would, if your not testing, reviewing and ensuring your code is solid then I don't care if an ai wrote it or you wrote it, it's not going in.
But other people are different.
SoCuteShibe@reddit
It's exactly that. Tons of ppl in this thread are getting it wrong.
blakfeld@reddit
Depends on where you work. I’m a staff engineer at a company you’ve heard of, and if I don’t primarily vibe code I would almost certainly be on the fast track to a PIP. We’re literally having a big company trip to get everyone together to “teach vibe coding”
golden_eel_words@reddit
Same here. NOT leveraging these tools is currently seen as a big red flag for engineers. I don't necessarily like or agree with that (it's incredibly nuanced), but it's the reality.
blakfeld@reddit
Well said. I’m not a fan of what the futures looking like, but I do like eating regularly, so I guess I’m learning to vibe code
mouse_8b@reddit
I admit there likely are senior devs who would frown upon it.
Kinglink@reddit
At least as I understand it .. that's not vibe coding. Vibe coding is most hands off the wheel just trusting the code to be correct or at least minimal testing.
Your final line tells me you are definitely not ;)
cplusequals@reddit
Vibe coding is just heavily leaning on AI. There's no difference in how its tested.
SoCuteShibe@reddit
No I think King has it right. That's where the 'vibe' aspect comes from. You don't know how it works, it's just a vibe thing.
TDSGoldenSun@reddit
I've never seen a company low their testing standards just because some of the code was written by an AI. This doesn't stand up to any amount of scrutiny in the industry. Seems like a Reddit thing.
SoCuteShibe@reddit
I'm not sure I'm getting your point. You seem aggressively committed to giving the wrong interpretation of the term "vibe coding" credence.
Vibe coding is something that beginners, amateurs, wannabes, bad students, etc. are doing. Anyone who is employed and vibe coding is almost certainly an intern or first in line for a layoff.
Remember when Stable Diffusion came out and eveyone was suddenly an artist/prompt engineer? Think that, but for coding, and hacking together simple apps. That is what vibe coding means.
People that think vibe coding means using AI tools are just totally not getting the point of the term.
Source: industry professional/certified anti-vibe coder
TDSGoldenSun@reddit
Aggressive? The opposite. This sub is aggressively opposed to using AI as a tool to assist their development. I promise you, the people that are oppositional to AI are going to fall behind those of us that are intellectually curious and honestly pursuing tooling that maximizes our productivity. The proof is in this very thread. The second you mention you use AI tools to deal with the boilerplate shit juniors shouldn't have to touch you get accused of not testing or "prompt and ship." Frankly, it's a self-report. If you haven't been able to find the utility in AI at this point in its maturity either you're insecure about your ability to be productive or you haven't been keeping up in the field.
Source: Experienced industry professional/pragmatic developer involved in hiring.
TehTuringMachine@reddit
Using AI for boilerplate code or setting up code isn't vibe coding, that is the sticking point. If you put any brain power into analyzing, testing, or otherwise thinking about your code, you aren't vibe coding.
mouse_8b@reddit
I got a little off track in this discussion, and some of this thread got into code review processes at jobs, but I did want to revisit the question.
In a hobby project, not in the context of turning something in or getting a review, I have used JetBrain's Junie to vibe code past the boring stuff. And then yes, I just run it. And if it doesn't work, I clarify the issue and try again. I was able to make significant progress on an app I had stalled out on in just a few vibe sessions.
Of course, it can only go so far. Yes, I had to dive into code that I didn't write and figure out a bug. That's a huge part of my actual job though. I'm really good at that. In a way, that's easier than a green field project.
Do I have to stop saying that I vibe-coded this project because I opened the editor?
Is "vibe-coding" to the point where it is a drop-in replacement for a developer on a professional team? No.
Can we learn anything about software development with "vibe coding" techniques? I think so.
Do I think "vibe coding" is the wave of the future. Yes. I wouldn't have thought so until my experience with Junie. I've used CoPilot for almost a year, but Junie is in another league. You really can vibe code past the easy parts and focus on the actual puzzles that make coding fun.
cplusequals@reddit
That's not a serious question. I obviously can't answer for him, but in all my years as a software engineer I would never presume to ask such an insulting question even to an intern. I give my fellow developer enough grace not to accuse him of not testing his code locally and using basic CI/CD. That doesn't always pan out as some developers have had unique career paths at small companies that don't have industry standard practices, but you're going to be wrong far more often than not if you assume the worst.
Kinglink@reddit
Buddy you don't know why I asked so next time keep your opinion to yourself.
cplusequals@reddit
Likewise. Stop being an ass. It's not appropriate public conduct.
Kinglink@reddit
Look at your downvotes and realize that because how you've acted. Next time try a little maturity.
Don't bother trying to reply. Your cut off here
TDSGoldenSun@reddit
I don't know why you'd stake your argument on the reputation of people that frequent this sub. This is one of the worst forums for competent software developers. It's very obvious the majority of people here are insecure about AI taking their jobs and the fact that you'd presume someone using AI isn't going to test their code is proof positive of that.
I learned very early on that the very worst people to work with are those that feel as if they have something to prove. It's an evergreen lesson you might want to heed.
FlyingBishop@reddit
I never check anything in to main without reading and understanding it. But I find myself running vibe coded stuff increasingly often.
TDSGoldenSun@reddit
Are you honestly asking if he's testing his code locally? You can't be serious, can you?
g1rlchild@reddit
I'm sorry to whoever wrote this, but it feels like an AI could have written it. It just articulates obvious points that anyone aware of the issues involved already knows.
wRAR_@reddit
It's a usual sel-promotion blogspam account, of course the article is AI-written.
iamapizza@reddit
Also so many em dashes.
shevy-java@reddit
Like tabs versus spaces, right? The tab users are wrong!
Blanglegorph@reddit
You better believe I checked whether you used the right character.
VibeCoderMcSwaggins@reddit
https://github.com/The-Obstacle-Is-The-Way/clarity-loop-backend
Vibe coded all this
I do care about mastering coding Both natively and agentically
ixid@reddit
Exactly this, vibe coding is amazing as a less technical person for getting work tools from an idea to something that produces value. We care about the outcomes of the tool, not the coding that produces it, and if the quality becomes an issue we've essentially made a prototype for a real developer to turn it into something more solid.
sayris@reddit
No idea why you’re being downvoted, at my company we follow a procedure where the prototype is always thrown out and recoded when we’ve established our value proposition and got stakeholders interested
If vibe-coding can get to a working prototype phase twice as fast (that we’re going to throw out anyway), then what’s the big deal if the code isn’t super scalable or production ready right now
ixid@reddit
I think talking positively about vibe coding is just unacceptable to the purists, but who knows, the reddit hive mind is pretty random about why it decides it hates posts.
XpanderTN@reddit
I'm a self taught software engineer. I'm as far from a purist as I can be, and I agree with them. There be dragons, but you know fck us for mentioning why.
Kuinox@reddit
No no, it's not pretty random, reddit is filled with anti ai luddites.
PaintItPurple@reddit
Yeah, I bet these luddites said the same shit about NFTs and the metaverse, and look how that turned out.
Oh, wait
Kuinox@reddit
No they said that digital art, using a computer to draw, wasn't real art.
overtorqd@reddit
I, for one, appreciate the willingness to have a balanced conversation. All I hear about AI coding is LinkedIn and business articles praising it as a magic elixir that eliminates the need for engineering, and deeply offended Reddit developers who think its a useless toy.
The sales and non technical leaders will never understand the engineering side. We cant let them own the narrative, but we shouldn't put blinders on either.
dalittle@reddit
Ever heard of the netscape re-write?
https://www.joelonsoftware.com/2000/04/06/things-you-should-never-do-part-i/
I am guessing maybe you have some niche where this works, but doing this for an actual product in production prototype or not does not always go well for all the still relevant reasons that happened to netscape.
crazy0ne@reddit
I think you are referring to a vertical slice prototype, something that can show a representation of a product to gain shareholder interest/confidence.
Your support of comment is invalidated by the fact you know you are growing out the vibe coded prototype, and the original comment's claim is that you can evolve the vibe prototype. Two very different claims.
ixid@reddit
I didn't say that.
crazy0ne@reddit
I think you did. Maybe you meant something else, but it is also what you said.
ixid@reddit
No, I didn't specify what happened to the prototype in that comment, you just assumed, and in another comment in this thread I said you replace the prototype entirely with proper code. It's far better to do a clean implementation than build on a prototype, whether an AI or a human wrote it.
crazy0ne@reddit
I get and can appreciate that.
Let's just toss it up to tech debt since you have some comments to clean up. 👍
lookmeat@reddit
One day our children will curse vibe code the way we cursed slap dashed PHP and JavaScript websites.
You aren't building the prototype, you are building the fuse version and then there will never be time to sit down to turn it into something more solid, and it'll never be fine but a "real developer", you'll be the one who will find yourself getting hired to maintain it.
ixid@reddit
If it makes more money than the cost of the developer what's the problem? We're obviously not talking about safety critical software here, just business logic and related tools, which already mostly exist in a hellish state of half-broken Excel sheets that no one understands.
XpanderTN@reddit
If you care about scale and maintenance long term, then you know the problem.
Engineering and its principles exist for a reason, and its to prevent or reduce problems already seen.
You can cheap out and think you'll not need an Engineer. You can suggest that it'll be 'easier' because you have a working prototype, but you frankly don't know that. The way you build a working prototype architecturally could be pretty bad and lock you into a situation that requires a full refactoring of the code for it to be scalable.
If you want one off tools to solve small problems, whatever, but actual applications that provide value? No.
Business Analysis and business in general has the concept of Non-Functional requirements, and if you don't already have a knowledge for what those are, your working app is just a ticking time bomb that looks like it works.
FeepingCreature@reddit
Yeah but people are wrong about PHP and JS as well.
lookmeat@reddit
I think you can have good quality PHP and JS. I think there's great products that show you can totally scale it (e.g. Facebook). But you need to do it well, with good engineering and understanding. They aren't shortcuts, just different tools with different pros and cons, you still need to do the hard part.
Same thing with vibe coding. I do believe that it's a viable and solid tool. There needs to be improvements and better ways of using the tool, and it's exciting to see what we can or cannot do, right now I don't think we are using it as best it can be, so things will keep improving. But it needs to be well done. Not a shortcut either, you still need to do the hard part, and that needs work.
FeepingCreature@reddit
Yeah but doing it badly and falling on your face is still as always how you get good. Programming is all about frustration tolerance and frustration management, and the fact that there's languages where you can eschew style and, to some extent, skill and just get going is very good for that. JS and PHP have gotten a lot of people into programming, and AI will do the same.
Luolong@reddit
Don’t be surprised if the bill from real developers hits. There is a term called “technical debt” that is a real sneaky bitch.
ixid@reddit
I am very aware of technical debt. As I said that's why we hire real devs, who cannot be replaced by vibe coding.
exmachinalibertas@reddit
You shouldn't be getting so heavily downvoted. This is a valid take. I hate that it's the world we live in, and that society incentivizes business to act this way, but it's a totally valid business reason given that we do in fact live in this society and this world.
HomeyKrogerSage@reddit
I don't think you understand the term technical debt. That's debt. Like owed money. Money you wouldn't owe if you had just gotten a developer in the first place
overtorqd@reddit
People take on debt all the time. A mortgage lets you actually have a house before you can afford to buy it outright. It's like saying a mortgage is money you wouldnt owe if you'd just bought the house with cash.
Technical debt isn't always bad. If it allows you to get a product to market and survive as a company, to get paying customers and revenue, I'll take some tech debt.
cake-day-on-feb-29@reddit
I mean if it's necessary then it's necessary, though I'd question how you got to a place where it's AI vibe coding or die.
Because AI tech debt isn't just any tech debt, the code it produces is barely coherent, at least tech debt from a human is somewhat consistent(ly bad).
(Not to mention the potential legal debt if/when the courts decide you can't launder code through an LLM to get rid of its license)
ixid@reddit
I do understand technical debt and in this context the alternative scenario is no product or fewer products.
HomeyKrogerSage@reddit
How? Your logic just doesn't flow. Either you have technical debt and don't have a viable product or you foot the overhead to get a proper developer and actually have a viable product like I don't see what other train of thought you're following
ixid@reddit
You can make a janky AI prototype that checks it the idea has the potential to warrant getting a dev and doing it properly.
epicfail1994@reddit
I mean, you usually try to avoid tech debt in the first place
You literally sound like you have no idea what you’re talking about
crazy0ne@reddit
This is like saying:
"That is why my house has running water, so I can quickly and effectively put out the fires I lite in the living room to control the burning and still keep the house warm without a heater!"
The water and the heater serve different purposes that prevent the need to start a fire in the first place.
archlucarda@reddit
that last part invalidates the rest. a developer would have to ctrl-a + del AI slop to get a system functional to any meaningful kind of sophistication. this is the sort of fundamental to software engineering that LLM advocates choose to ignore -- the design decisions you make early on are practically 90% of the gig. if someone hands you several hundred lines of code that work 70% of the time, kind of, you are not ever going to be like "oh, well thankfully i just need to do the 30% remaining work", you are going to be like "jfc what idiot wrote this garbage, it's all wrong, it'd be quicker to just start over jfc"
ixid@reddit
Yes, it's pretty normal to entirely throw out the prototype and code it properly once you know the value proposition works. AI code isn't good or safe or reliable, it's just quick.
mfizzled@reddit
we actually have a policy of always throwing out the prototype code once it's been a useful poc
ivancea@reddit
But the dev now has a real working prototype instead of a badly described task. Communication with clients is a real problem, and this case solves a big part of it.
Not only that, but the prototype may be usable and easily fixable. As a real example, I asked my partner to vibe code a website he wanted. He's not technical, at all, he's a hairdresser. He did it, and the website worked very well in a few iterations.
From behind, I was checking the code that was generated. It was correct, it was what you could expect, and easy to follow. Claude 3.7 Sonnet.
Does that mean that all vibe coded code is like this? Of course not. This is the best case. And it's a common occurrence. So it indeed helps people create real projects, and if a programmer is needed, they have a good chunk already made. So it will always be cheaper (just for not having to fight with the textual definition alone)
FeepingCreature@reddit
You can of course tell the AI to refactor in certain ways.
telewebb@reddit
Damn dude, you got downvoted for having a good time.
ixid@reddit
It's my most downvoted comment ever, I feel pride.
zrooda@reddit
Over the decades I've seen many beautifully architected codebases die on the simple problem that the app was in fact shit, and I've seen amateur-hour projects made of spit and lube become a serious thing. At the end of the day the target audience doen't care about the underlying solution outside of a horror leaking out into tangibly bad bugs or UX. The inconvenient truth, nobody cares when it works and design always was king.
Vibe coding doesn't die on "not understanding" implementation details but on current-gen AI hallucinating problems it can't fix and driving itself into a wall long before project maturity. If these are ironed out to a point where a muddle with an idea can actually ship a somewhat working app, the craft and the market will do an immediate 180. Vibe coding is only an illusion as long as it doesn't really work and it doesn't need to be perfect to work - spit and lube functional is enough.
Whether it comes to that point is really Pascal's wager on our end. If it doesn't and vibe coding peaks at producing useless concepts, we can pat our backs about the educated scepticism. But if it does, how we feel about the nuances and "mastery" of the craft will be a dead proposition - nobody will care, not the users, not the market. Maybe some traditional "hand coding" shops will pop up in Tokyo.
IMO the reasonable thing to plan for is not overfocusing on being an even better programmer but diversifying into the satellite areas of product design, consultancy or project management - parts of the trade where ideamaking and educated decisions will remain to matter. The market is salivating to shed pure developers who imagine architectures and write their code.
If you make yourself more flexible, the worst that can happen is having more holistic experience traded for forgetting how to reverse a b-tree that you've never had to do yet outside of a job interview anyway. If you don't, the worst that can happen is your entire complex mastery of the craft vaporizing into thin air, choked out by no demand.
greenknight@reddit
When this was a bigger part of my life I started TAOCP but never finished (in my defence Knuth hasn't finished his end either) but I'm looking for something in that space again.
Any suggestions?
zrooda@reddit
https://en.wikipedia.org/wiki/The_Design_of_Everyday_Things is a must read, I also liked https://universalprinciplesofdesign.com/books
FeepingCreature@reddit
There are no shortcuts to mastery, but there are now amazing shortcuts to:
Vlyn@reddit
Yeah, I dunno. I was just warming up to using AI instead of StackOverflow and docs at work.
Went well at first, until I had a more fringe problem with EFCore in C#. ChatGPT happily gave great answers, but each one of them had at least one hallucination.
Like a "GetModelDifferences" function, which was exactly what I needed, but straight up doesn't exist.
When telling the model this function doesn't exist it will apologize, say it was incorrect, there is no such function and offer a different solution.. with the same issue.
Super frustrating and in the end I wasted a lot of time.
FeepingCreature@reddit
ChatGPT is kinda mid for coding. Try Claude 4. If a function doesn't exist, it can go look up the docs on its own.
Vlyn@reddit
Thanks for the hint, Claude 4 did quite a bit better with my question. I'll have to check if we have it available at work :)
IAmTaka_VG@reddit
15k loc is honestly a lot of apps on the app stores.
Vibe coding may never fly in enterprise monoliths that are 200-500k lines of code but that doesn’t mean vibe coding isn’t about to destroy a trillion dollar industry.
rnicoll@reddit
So here's what I don't understand; if anyone can make a 15k LoC app from an idea, who is buying an app any more?
Surely the friction to recreate the app from its idea is what people are paying to go around?
shevy-java@reddit
May depend on the time; if you earn 100 Euro per hour, then purchasing an app for 5 Euro is almost irrelevant, if there is time saving to be had for your use case.
arcandor@reddit
I think two things are happening at the same time.
AI can be a force multiplier for programmers who know what they are doing and also take the time to learn how to leverage these new tools and models most effectively.
Also, the bar for 'idea to somewhat working code' has been dramatically lowered, allowing many more people to get to this 'level' of sophistication. They could never get any better, or that could be their entry point into the profession.
WriteCodeBroh@reddit
I’d argue they really can’t. Anyone can say “I want XYZ,” great. Unless it’s coming out of there perfectly, good luck getting everything working if you are just a “vibe coder.” I’d argue that engineers even prompt the LLMs better, which results in better output to begin with.
Or you know, even simple things like, can a vibe coder bundle up their Node API? Will they be able to deploy that to AWS correctly? What will they do when something breaks? How will they scale it?
Or do they just do local mobile apps? What happens when they have vulnerabilities they need to patch and release updates? Will they know/care to do that? If they do, what happens when a dependency update breaks their app?
Just off the top of my head.
FeepingCreature@reddit
Yep! So this is bad for developers but very good for the smartphone as a device.
You just tell it what you want to do! Eventually, apps will completely disappear as a concept, there'll only be libraries and components assembled on demand. The premise of OOP finally made real, and all it took was AGI.
VLHACS@reddit
I'd like to think that AI tools such as github Copilot as simply a shortcut between encountering a problem, looking up stackoverflow articles, googling random forums and technical documents to gain a small understanding of WHY something was broken, put in the fix, and move on. Instead Copilot summarizes an issue for me given the context of my code (huge benefit from the former process), and provides an example of how to fix the issue. I then decide whether to use that snippet or not.
I rarely use code completion unless it's for documentation or filling in rote entries
Vlyn@reddit
Yeah, I dunno. I was just warming up to using AI instead of StackOverflow and docs at work.
Went well at first, until I had a more fringe problem with EFCore in C#. ChatGPT happily gave great answers, but each one of them had at least one hallucination.
Like a "GetModelDifferences" function, which was exactly what I needed, but straight up doesn't exist.
When telling the model this function doesn't exist it will apologize, say it was incorrect, there is no such function and offer a different solution.. with the same issue.
Super frustrating and in the end I wasted a lot of time.
shevy-java@reddit
The sad thing is: Google search really sucks now. I had to adjust my old strategies - searching for stuff often does not work anymore (at the least not via google search). I try to gather useful bits more locally and try to get people to also store more stuff in e. g. github repositories directly, so that people can find it there.
XpanderTN@reddit
This is how I see it too. Compresses the time you would already take doing archeology or research.
Still have to know the right questions to ask. It's always about the question.
corsario_ll@reddit
I'm software engineer forma decade , the IA just make the core problems of software development (scalability )worst
chat-lu@reddit
One of the core problem is the Church of scalability. You aren’t the next Google.
rnicoll@reddit
I mean some of us are the current Google.
shevy-java@reddit
It's that bad already?
rnicoll@reddit
I was making the point that some of us do need to engineer for big tech scale, because we actually work at big tech, as a counterpoint to the comment that "you aren't the next Google"
5scap5@reddit
It scales bugs faster for sure. :)
Empanatacion@reddit
Can we collectively decide to start mocking people that use the phrase vibe coding?
shevy-java@reddit
It's just with the word hipster years before.
If vibe is now the bad word, hipster may have gotten an indirect upgrade, such as in:
"In this image, hipster Joe is using AI to write more vibe-approved code."
florinp@reddit
thanks god we have another hype word: vibe.
That's always in need in IT: hype. It looks that hype is like gas for cars.
Yangoose@reddit
Remember when everything was "machine learning"?
shevy-java@reddit
It got an upgrade to AI!
shevy-java@reddit
I think we had this in the past too, just take MTV back in the days and music memes - Vanilla Ice Ice Baby, Rap-culture and Hip-Hop culture (those strange wide trousers e. g. in songs such as "can't touch this") etc...
If music has fashionable words, why not in programming too. All those vibe hipster hackers!
SophiaKittyKat@reddit
I just wish people would stop pinging me on PRs they obviously haven't even read themselves, and expect me to review 1000 lines of some completely new vibe coded feature that isn't even passing CI. You can figure out how to merge that shit without attaching me to it.
clbustos@reddit
That happens to me, but in the context of statistical assistance. A client came to us with pages and pages of R code from ChatGPT and wanted it reviewed. I said nope and delegated it to someone else.
shevy-java@reddit
That's actually interesting, though, because real humans do the quality-control step. Obviously AI is still lacking if it needs humans to do quality-control steps.
gazofnaz@reddit
I got pinged to review an RFC document today - from a senior DBA.
The document was clearly lifted straight from GPT, which wasn't a surprise.
I added some comments, in a professional tone, along the lines of, : "we can't fucking do this. It's way too much fucking work and doesn't even fix the fucking problem at hand."
I swear to god, he pasted my comment in to GPT and pasted the GPT response right back. And he's obviously using that version that had the hysterically over-friendly vibe going on; "Wow, yes, what a fantastic detailed and insightful comment. You raise some excellent points."
I'm basically working with a Human AI Agent. And our database is still fucked. And I'm still going to be the one who has to fix it.
Yangoose@reddit
Meanwhile this ass hole is bragging to his friends at how he's just using AI to do his job and nobody even notices...
itsgreater9000@reddit
Had a more senior coworker act like I was supposed to be impressed after having used AI for the past 6 months to do his work... it became quite clear after his "productivity" increased so quickly... and let's just say prod has never had more bugs...
Outside-Chemistry180@reddit
u mean his script cheat in random 2d io games or what?
jimmux@reddit
It's funny how we got used to LLMs using that tone, but when it comes from a supposedly real person it sounds so very sarcastic.
Have the machines been condescendingly mocking us the whole time?
bcgroom@reddit
Our CTO did this recently to me, it didn’t even compile 🤦♂️
chat-lu@reddit
Merge it. Prod may break now, but you probably won’t have such PRs in the future.
PeachScary413@reddit
Have fun being the "guy who broke production" and getting all the blame for not properly reviewing the PR 😊👍
exmachinalibertas@reddit
"The AI reviewed it and said it looked good."
-grok@reddit
lol don't touch the poop
retro_grave@reddit
Oh sweet summer child. You think people sending these PRs learn anything? You stamp it and you'll be expected to fix it. Hard no from me dawg.
BoilerEuler@reddit
I know it's real, but I still can't wrap my head around people actually doing that. It just feels so far below the minimum bar of professionalism, to actually understand what you're doing. It's like if an electrician just threw a bunch of cables through your walls and hoped it all worked out, instead of running them to specific places with intention. I feel like in any other industry, it would be patently obvious when people were fucking up on a similar scale.
PeachScary413@reddit
They are doing it for their CVs, so they can put "Frequent contributor to Project X" on them.
exmachinalibertas@reddit
I just last week asked a principal why he added a line of code that seemed to not do anything, and he replied "I don't know, the AI told me to do it".
I'm very salty he's a principal. And yes, people are actually doing this.
DeadlyGlasses@reddit
And the same people expect to get paid like 2-3X the other profession for some reason.
rnicoll@reddit
> just threw a bunch of cables through your walls and hoped it all worked out, instead of running them to specific places with intention.
I feel like I've lived in rental properties like that...
Outside-Chemistry180@reddit
me too, i hope that vibe coder start check their work lmfao.
i think exist some type coder.
1. vibe coder
2. coder using llms after google and yts
Yaoel@reddit
I fired someone Monday for doing exactly that, made an example out of him for the rest of the team but he was already on thin ice
axehammer28@reddit
Holy shit this would make me lose my mind
owogwbbwgbrwbr@reddit
In what context does this happen? How is the work structured to just allow random PR assignment?
FeepingCreature@reddit
Github.
VibeCoderMcSwaggins@reddit
Vibe coded all this
Chat with your apple health kit backend Trying to build out to a psychiatry digital twin
EEG-depression ML notebook built by me All agentically coded
Started coding on 2/11/25
https://github.com/The-Obstacle-Is-The-Way/clarity-ai-jupyter
https://github.com/The-Obstacle-Is-The-Way/clarity-loop-backend
But I really need a real devops engineer to see if it’s okay.
Is it slop? BS? All shit?
Ikinoki@reddit
There's no illusion of vibecoding. What I found is that models' performance is getting forcibly degraded by the operators!
I don't have solid proof at the moment as I wasn't tracking it, but in my last response out of 4 tested systems only GPT 4.5 output correct (not perfect) response https://www.reddit.com/r/LocalLLaMA/comments/1kz5onp/one_shot_script_conversion_from_shell_to_python/ .
People said that it is wrong to expect one shot, however this never happened 1 year ago with poorer gpt4. The quality of code was good, it didn't understand correctly that's true but I check earlier projects I used GPT for and the quality is pretty stellar, not single time did it try to loop over words in python.
What's surprising is that even older llama 3 3B models outperform the current top notch programming models.
pythosynthesis@reddit
Vibe coding will turn out to be a blessing for the next gen of coders. Not because we'll all become love coders, the opposite. So much garbage shit will be produced by a wave of MBA wanna be vibe coders that systems will start to crack at the seams. More vive coding won't solve the problem and a different breed of MBAs will realize they need actual coders. Quite possibly with a return to really old school, low level skills. Those who will know will get seriously good salaries.
Something similar already happened with early internet coding. People decided coding wasn't such a big deal after all, they could put some HTML and php here and there. Shit stopped working real fast.
Unfortunately the price to get to El Dorado is the current career dread cum garbage job market. All of this peppered with talks by underdeveloped management (non-technical) consultants who will force upon us the mantras of vibe coding with "everyone can be a coder" chants in the background.
Stay strong fellow engineers, this too shall pass. The road to El Dorado is paved with vomit. The strong will enjoy the spoils.
gioraffe32@reddit
A friend and I were discussing this the other day. Neither of us are programmers. We're both IT, but I'm more Sysadmin, while he's a Business Analyst/PM. While my major was CS, I never finished. His background is IE.
Lately, he got an idea to code a POS system for restaurants. He knows some restaurant owners who were telling him how expensive these systems could be, and how there's some other small local business that's started providing their own proprietary POS service that's cheaper than the big players. Yet this company is apparently making money hand over fist.
While he's dabbled in programming before, he doesn't have any foundational knowledge. So to create this POS, he's been vibe coding. It's impressive what he's been able to create in such little time. It's a simple web-based app. I mean, a POS should be simple to use. Regardless, it's a cool little thing.
As he was raving about how much he can do with LLMs and how good they are these days, I asked him, "Is it secure?" After all, it'd be dealing with money and credit cards and all that. Sure, his idea is to integrate with a payment processor like Square or Stripe, that handles the actual processing of credit cards. That makes total sense. But is his program knowingly or unknowingly storing that credit card information? Are the variable being properly scoped and such? Does he even know about that?
There are so many more questions to be asked, questions that I don't even know to ask, because I don't have fully foundational knowledge, either. But if one doesn't have that foundational knowledge at all, how can they be expected to ask anything, much less answer them?
He's having to read the generated code, parse it, and add his own code to connect and fix problems; I can't imagine the LLM has literally created an entire working program for him. So he does have to understand it at some level. But if he doesn't know what to look for, he's going to miss all the things that may work, but actually are problems. It's not enough that the program just works.
I agree that vibe coding can be super helpful. Like other's have said, it works for simple scripting, POCs/prototyping, or simple internal tools. I have a coworker who's vibe coding something akin to Uptime Kuma for our team. Cool, I get it. Vibe code away, my guy.
But a POS system? That's a whole different beast. I wouldn't be comfortable using my friend's system. Is this just a proof of concept or is this something that he would actually be willing to "ship?" I certainly hope not the latter. Though he didn't say it was just a POC. But again, he lacks that foundational knowledge. Who's to say he wouldn't try to find a restaurant to work with?
Honestly, I don't know how serious he is about this idea. But I hope our discussion gave him some additional things to think about if he is thinking seriously about this POS idea. He's an intelligent guy, but sometimes I think he thinks he's more adept at tech than he really is. And I think for people like him, vibe coding can greatly feed that notion.
At the end of the day, LLMs are just another tool. And like any tool, you have to know when and where to use it.
LessonStudio@reddit
Most of the time the Copilot autocomplete is the next 4-6 lines of what I am about to type. But, when it comes up with 40 lines, it is usually very wrong.
Often what it suggests is almost exactly how I would have formatted it, and how I would have structured it. But, once in a while it does something interesting and cool.
I see this as a kind of pair programming.
Other than the longer nonsense it suggests, my main problem is that it often codes in an older style or uses an older API, etc.
carrottread@reddit
Judging by amount of text decoration in this I think it was Vibe Blogged.
anshou@reddit
vibe coding is what happens at the edges of the bell curve while the engineering is in the middle
visor841@reddit
At first I read that as "there are no shortcuts to misery" and was thinking "there are tons of shortcuts to misery, I've found many of them".
weggles@reddit
Someone high up at my company said a Jr can go to Sr in 18 months with AI
bzbub2@reddit
while the article is true, it's doesnt really say it in a very powerful way. like, make me believe. I bet most up votes here are just vibevotes
Berkyjay@reddit
It's fun sitting here watching these two sides argue about this. One side are the people who thought there skills were something special and so felt that they were special as a result. The other side are people too lazy and not at all interested in producing a good product. They just want to put something out there, take the credit (money) and move onto the next hussle.
Meanwhile, the majority of us are just trying to keep our jobs and stay relevant. That's going to mean adopting these new tools. I for one welcome coding assistants and am enjoying how much more productive I am with them. But I also really care about what I produce and I take pride and responsibility in it. So if an AI gives me buggy code and I don't spot it, then that is 100% on me.
_insomagent@reddit
Photoshop was the end of graphic design, amirite?
Skizm@reddit
To mastery? No. To productivity? Yes. If people are successful vibe coding, then more power to them. I've just not seen any serious work done with it yet. I've definitely seen people spin up MVPs via vibe coding and extract VC money based on smoke and mirrors, but that's nothing new. Just different techniques to get a jank, barely functional, demo through a pitch meeting.
azhder@reddit
People are already highly productive if it comes to producing shit.
levodelellis@reddit
Hold up, people think vibe coding is real? lol
TrekkiMonstr@reddit
This reminds me of those articles that always pop up about studies showing that study drugs (Adderall, Ritalin, and generics thereof, for the most part) don't actually make you any smarter. Of course they don't! That's not the point -- the point is to make it easier to study more, not better. So, "Adderall won't make you any smarter" falls on deaf ears, just like "that fancy car won't get you across that lake" would. And it seems similar here. Sure, there are no shortcuts to mastery in chess, but there is absolutely a shortcut to beating Magnus Carlsen. It's called Stockfish. Or any of its friends.
Now, is AI at that point yet? Absolutely not, no question. But then, that's the argument to make -- not that it fails to make you better, but that it fails to do what you do want it to do.
Kinglink@reddit
Amen...
Every time I hear "Vibe coding" I cringe. I also cringe when I hear people say "Ai is worthless" because it's not.
The thing is it reminds me for how long I was a "Senior engineer" but had no understanding what that means. If you're a Senior and only work by yourself... you're not a "Senior". If you only work with other Seniors... you're probably not a senior either. If you never write documents, you're definitely not a senior.
Senior engineers design more than they code. Honestly all levels should be doing some design/architecture work, but at a senior level you're doing code reviews, you're analyzing big problems for others, you're dealing with the hard problems that need to be explained. You're guiding the ship (or a part of the ship).
The thing to become a senior you need to read and learn how to code. Vibe code is... like it's what a Junior does. Write code that barely works and a senior comes in to explain why it doesn't. The Juniors that don't do that, quickly get promoted, the juniors who continually do that.. well...
The thing is for AI, you need to be the senior programmer, you need to define what you want the AI to do, and confirm it does it. "Man that's tedious and boring". Welcome to the world of a Senior Engineer.
The point is not to say "Well don't use AI" no, we all should be using AI some of the time (possibly most of the time) but if you're going to be a programmer with AI you need to be a good programmer and guide the AI, rather than give it some simple command "make me a game".
Someone who defines "I want a game" Is going to get something crappy. Someone who defines "I want a RPG where characters are defined as XYZ. The story is based on this script I've generated, the gameplay is setup like ABC..." Is going to hopefully get something really good (or really they'll have to break down that prompt further.. which is the point).
Really it's "Good input in, good output out." That's how it works with Junior Programmers, that's how it works with AI.
Basically puts on the old senior engineer hat CHECK YOUR AI'S OUTPUT YOU DUMB FUCKS...
LucasOFF@reddit
Quick MVP/proof of concept using AI tools? Sure. Genuinely good quality code, especially if it was prompted by someone inexperienced? Fuck no, it's a disaster. We just have to ride out this hype wave before we're back to the routine work. It's like companies will make 99 mistakes before making the right decision, replacing people with AI - is one of the mistakes.