The article itself doesn't have comments. On HN, there's 2 posts that are against vibe coding which could be seen as gatekeeping. What are you referring to when you say "I see many gatekeepers in the comments."?

there are now 6 comments all critical of the notion of vibe coding, this thread being the only one started by someone who praises its benefits

Hey, that could be considered many. 2 many.

some people see what they want to see

I don’t see anyone “gatekeeping” but I do see warnings. I don’t know what you’re good at, to give a good example. But if you have a hobby or job, imagine the new guy coming in, knowing absolutely nothing and doing shit they technically can do, but shouldn’t be done. That’s what this is. Just like a new cashier can technically just not give correct change, everyone knows this will end badly; eventually, even though everything looks fine from the outside.

That’s what AI code tends to do. It looks fine to the outsiders, it might even function correctly MOST of the time. Programmers are trained, mostly through experience but also classically, to think through every exceptional case and figure out how to handle it (or return an error so a human can handle it).

That isn’t to say this is ALWAYS the case with AI code, but rather it TENDS to be the case. YMMV, which is where these warnings are coming from.

> Programmers are trained, mostly through experience but also classically, to think through every exceptional case and figure out how to handle it (or return an error so a human can handle it).

Unfortunately, vibe coding is seen as acceptable to many specifically because what you say here hasn't been the norm for like 15 years now.

The rise of bootcamps, dependency-driven-development, and "move fast and break things" culture convinced a whole generation of programmers that all that matters is the happy path, and even then, only within the context of today's specific task. The ocean of garbage they produced during that time is both one of the reasons why LLM's can produce code in the first place and one of the reasons why that code is so consistently and irrecoverably poor by traditional standards. Aspiring vibe coders see their peers earning absurd six-figure salaries to produce the exact same sort of unstable, short-sighted noise that they see their LLM literally reproduce for pennies now.

In many ways, we should be gatekeeping by asserting a higher standard of foresight and quality for redistributed code, but we completely lost that battle many, many years ago already.

> It looks fine to the outsiders, it might even function correctly MOST of the time.

Most code (even written by good programmers) rarely functions correctly most of the time. Most code is broken. This is not the problem with AI. Unless I am using the tools wrong, LLMs can generate fully functioning scripts (and some of them are good) but they break after the 50k token context and start doing insane things that not even juniors will do (like randomly removing code).

If you want to see a shit-show, go to Bolt Discord channel. Some users are able to get a very simple and rough kinda single script app running. Everything else breaks once they start making simple amendments. This is not fixed by Claude 3.7 or O1 Pro or whatever. This is a fundamental issue in all of the LLM and a local maxima of the current tech.

Not that the current tech is not amazing. It is and there is a lot of value to be extracted from it. But everyone and his investor think they are about to reach nirvana and want to replace everything with "AI" where "AI" is a 100k context LLM.

I feel comfortable calling a lot of the comments gatekeeping "adjacent".

When a post is a "warning" about user's credit cards etc. and it was in none of the "vibe coding" examples, that feels like someone deflecting (but with a gate keeping mindset).

As others have said here, I too have knocked out little apps and sites vibin'. At first (1) it was to see if what everyone was saying about LLMs was true. Then (2) I wanted to see if LLMs could help in languages I was not familiar with. Then, even for languages I knew well enough, I (3) wanted to see an LLM's code to get the modern method of modularization for the language (JavaScript is one that has gone through phases). Finally I came to trust the machine and (4) have vibed code just for my own small projects.

Because people who admit they don't understand coding should not be coding with important data. And it's very easy for you app to suddenly have important data. Do you have auth? You have PII.

Gatekeeping is restricting access to the trade with arbitrary precedence, like "real coders use emacs." It's not expecting safety. If someone wants to build a bridge out of cotton the city won't let them. That isn't what most people consider gatekeeping.

What's stopping said people from grafting together a few Stack Overflow answers today?

Grafting together stack overflow answers requires knowledge of how your chosen programming language and execution environment work. There is at least some floor of knowledge required to synthesize those answers into a working solution.

Vibe coders as described in the OP can just copy/paste exactly what the LLM spits out with much less knowledge. It’s not the same thing at all IMO.

I don't see a major difference in a 20% working knowledge of a language vs 0% if the output works. Neither are writing provably correct code, and that's often fine! Many of the engineers at my company (not a software company) write terrible terrible code using the Stack Overflow method, but our manufacturing line still runs.

20% knowledge could be enough to fix errors yourself. E.g. this function name has a typo, or this calculation looks wrong, or this code has been repeated verbatim 40 times, maybe it should be a function, or why did the LLM change this totally unrelated area of the code?

If you only take exactly what the LLM spits out you can only verify by running. If it is broken in edge cases you can only prompt the LLM to fix. You then have to run the code again to verify the fix, and with zero understanding it may have broken some other edge case (LLMs do this all the time especially overly eager reasoning models).

> 20% knowledge could be enough to fix errors yourself.

It might be, it might not. My colleagues and I are also guilty of introducing unrelated edge cases, but again, we do fine.

The case you're describing is also fairly pathological. There's nothing stopping a vibe coder from digging into their code and becoming a 20% monster like me.

What you said applies to programmers who are actually good.

Consider that the world is full of programmers who are objectively shit (I live in India, and have personal experience). The only difference is that they’ll be copy pasting from ChatGPT instead of StackOverflow, and honestly I’d rather have ChatGPT.

So this is about the ego of professionals? As the other person explained, there are many use cases for disposable software, so it's a good thing that there's a new tech enabling non-programmer to create these on their own.

I never said there wasn’t a use case for disposable software. I only said I didn’t see gatekeeping here. Nobody is saying “don’t do it”. They are saying “be careful”.

We should also consider experts doing things that they know they shouldn't.

> if you have a hobby or job, imagine the new guy coming in, knowing absolutely nothing and doing shit they technically can do, but shouldn’t be done

Isn't that the definition of gatekeeping? and even so, who cares? the new guy coming in creating/doing something in a way that "shouldn't be done" will eventually fail and notice the shortcomings of what they did. Then they will either drop their project or learn the way "it should be done" (like we all did, except now with AI).

I kinda feel like this attitude, like “free market capitalism”, allows more bad things to happen with the potential for some unseen/immeasurable force to “correct” the environment once the Bad Thing is recognized and rationally dealt with. But there are probably better ways to avoid Bad Things that require only a little additional forethought, planning, and deciding ahead of time what sorts of things we’d like to avoid (oh no, regulation!).

Do we care? Is the podcast script timer from the article a Bad Thing? Are people actually hiring vibe coders for critical infrastructure?

I write really bad Haskell code sometimes without any AI assistance. Am I doing a Bad Thing when I create my own tools?

What are you intending to regulate here?

> Are people actually hiring vibe coders for critical infrastructure?

It appears to be happening right now in the US federal government.

To be clear, you're saying that people with no ability to reason about code are currently in a position to push unreviewed code into critical government infrastructure?

Do you have an example?

No. I never said you shouldn’t use it. I never even said it was a bad idea. But I did say to be careful. Find out if your cashier is giving the right change when you hire them, before you have the cops showing up to accuse your business of theft.

I'm not sure it's explicitly gatekeeping, more a lack of perspective.

The benefit and excitement is most felt by people with little to no experience writing code themselves. The fear seems to come from building.. what, code for critical infrastructure? That's just not what we're talking about here.

> I dont want them safety critical

Then hopefully they run on a computer without network access, otherwise everything is safety critical.

No gatekeeping. Anyone can write programs.

But to build something which could handle a customer’s credit card, password, or other PII and charge them for it, you better know what you’re doing.

It’s all fun and games until you’re the cause of someone’s identity or password getting stolen.

Anyone can use CAD software, but if you’re designing a public space, you better know something about safety.

To be fair, lots of "actual" programmers who don't know good from bad have been shipping insecure code to prod for decades.

AI is just another vector for this, not something entirely new.

When you have your amazing idea, instead of hiring an inexpensive low-skill developer (whose work you are also incapable of evaluating) to build and ship your idea in a low quality way, you're just paying AI to do it.

It's just putting they money into different (centralized) pockets.

No, people should not be knowingly half assing important things like PII just because "it's been done badly before." We make the good faith assumption that people who mess this up don't know better, not that they're willingly using a tool that will mess it up for them because they don't care.

I think you misunderstood my comment..

When a non-technical person hires an incompetent developer (that they likely don't know is incompetent at the time of hiring) to build something that turns out to be insecure - because the developer didn't know any better and the non-technical person doesn't have the skills to evaluate the output - no one was trying to do a bad thing, but they didn't know what they didn't know.

The non-technical person got something that did what they asked, without understanding all the underlying deficiencies.

It's the same with AI, I don't think non-technical people using AI are thinking "I don't care that this is building garbage code full of problems"..

Just like the first scenario, they don't know what they don't know, and they end up with something that does what they want, and that's a good outcome based on their limited knowledge.

To be clear, I don't think either of these scenarios is excusable or acceptable if you're working with PII or other security-sensitive things, I was just pointing out that this isn't new.

>Anyone can write programs.

Anyone can play the violin. Anyone can run a marathon. Anyone can …

People who spent their lifetime never quite able to sit down and write programs, for whatever reasons (time, focus, foundational knowledge, available mentors), have in the last year shipped working apps/scripts, by just saying in plain english what they wanted. That's exciting.

> I just want a free script or addon that will do its job instead of paying someone on upwork to do it for me. simple as.

So, you pay the billionaires to rent their graphics cards so you can avoid paying a normal person? -_-

If it is much cheaper, and the outcome is acceptable, probably even better, then that is what people should do.

I run my own model locally?

>The agents will be able to do maintainable code and safety critical code at some point in the future. It's not all gonna happen overnight.

Don't hold your breath. If the AI was good enough to do all that, "maintainability" would look very different from what it does today. What does it mean to be maintainable when an AI can completely rewrite the software in every iteration? If AI ever gets good enough to do all this, for safety-critical applications no less, probably 95% of the white collar jobs that exist today will be gone. There may also be robots to do 90% of all other work too.

What is important that it works for you. I think it is good that more users can solve their own problems.

There’s some gate keeping, but many of us are also worried about the inevitable future where we have to take over someone else’s “vibe coding” mess of code.

The vibe coding style translates to trying a lot of different prompts and small adjustments until it looks like it works. In the past these people copied from StackOverflow and poked at lines until it compiled and appeared to work, but that only gets you so far. Now those same people can bang away at an LLM assistant all day long and produce volumes of code that appear to kind of work.

I’m in another forum dedicated to programming careers. Every day there’s a new thread from someone asking how to deal with all of their junior employees spamming code review with obvious LLM generated code that they don’t even understand.

A lot of the defenses of vibe coding rely on the assumption that it’s in the hands of someone knowledgeable who only wants to save a little time for something inconsequential. That’s fine. What’s worrying is that vibe coding is being used as a replacement for understanding code for many juniors and lazy seniors across the industry as long as they think they can get away with it.

> ... many of us are also worried about the inevitable future where we have to take over someone else’s “vibe coding” mess of code.

Another potential alternative is that these things progress quickly and soon can code and review code at or above the level of most SWEs. I suspect that's driving a non-zero amount of anxiety in the comments.

I'm not an SWE, and frankly, they're above my level today. Saying "plz don't code if you can't understand it" applies to my code today without AI assistance. Are you suggesting I shouldn't release anything because others might need to read it? The way to prevent this within an organization is smart hiring practices, not restrictions on tool use.

> worried about the inevitable future where we have to take over someone else’s “vibe coding” mess of code.

worried? we're going to make a goddamned fortune

Doubtful. Usually when one person makes the proof of concept they do 20% of the work and collect 80% of the credit.

The people tasked with cleaning up the mess and making it usable in production do the remaining 80% of the work, but management is always disappointed that it’s going so slow relative to how quickly the proof of concept was created.

At one company it was widely known that the key to taking credit for technical work was to be the team presenting the GUI pieces to management. If you were working on something that couldn’t be shown to management with a “wow” factor, you were not valued. I see the same thing happening here, with vibe coders capturing the wow factor and the people who actually make it production stable being viewed as the slow and expendable ones.

People who made the initial thing in the absolute cheapest way possible expect the maintenance work to be similarly cheap. If it turns out it just can't be done on that budget, it'll often just not get done rather than pay what it would cost.

I think the issue is Hacker News users are so deep in the space that they're surrounded by AI noise at work every day and might even be forced to use it every day. So they've essentially accepted that it's a useful invention. There's this foundational, overwhelmingly positive assumption that goes unsaid, while 100% of their comment text is pure criticism and negativity. But in their brain, it comes across as maybe like 1% criticism and negativity and 99% positive lived experience. So they don't realize how bizarrely negative they're all sounding about a good thing. Of course, there are some people do really completely hate AI coding, but I don't think the general sentiment one gets from reading hacker news comments literally is accurate. It has to be read in context. At my work, people prefix their critical comments with "nit -" to reassure the reader that it's overall good.