Talk:Vibe coding

Example

Are we sure that the current (or previous) prompt really count as vibe coding? My understanding is that there is a distinction between using LLMs for coding (write a function to do x) and vibe coding (make me an app that does x). Write JavaScript code to estimate Pi using the Monte Carlo method. doesn't fit this to me, since it is stating a language and even a specific method to use. That's very different from Mirriam's definition: writing code, making web pages, or creating apps, by just telling an AI program what you want, and letting it create the product for you. I think we should update the example to be product-orientated. If we want to go meta, something like Make an app that counts the number of words in the Wikipedia article about vibe coding would be an option. SmartSE (talk) 10:45, 20 May 2025 (UTC)

Definitely not vibe coding - and lacking any reference to affirm that it is. I propose to delete this so-called example. Jmc (talk) 23:07, 23 June 2025 (UTC)

Nem con, I've gone ahead and deleted the "example". Jmc (talk) 21:03, 24 June 2025 (UTC)

There was a reference as part of the template, also, yes, a better example could be used, but the particular example was chosen for it's brevity. (Since LLM prompts can be fairly large) I oppose it's deletion since I think illustrating how vibe-coding is used in practice is important in the context of a thousand definitions being floating around. Sohom (talk) 03:00, 25 June 2025 (UTC)

There was nothing in that template reference to verify that it was an example of vibe coding. SmartSE has explained in detail above why it didn't cut the mustard as an example. I agree that illustrating how vibe-coding is used in practice would be nice, but I've not yet been able to find a RS with an example. Jmc (talk) 04:00, 25 June 2025 (UTC)

Just for fun, I took up SmartSE's suggestion to go meta and asked Claude to Make an app that counts the number of words in the Wikipedia article about vibe coding. It made a very presentable and functional app (counting 128 words) but at 280 LoC, it's rather too long to use as an example here. -- Jmc (talk) 04:30, 25 June 2025 (UTC)

There should be no requirement to have RS validate that something is indeed vibe-coded when it is literally a GPT generation of code. We don't ask for that level of proof from any other (technical or otherwise) article.

Regarding @Smartse's point of this "not being vibe coding", the snippet was added long before the dictionary definition applied the arbitrary gaurd-rails around the meaning, for a large portion of the time this word has been around, peeps have been using to refer to almost any AI-generated code as being "vibe-coded", which is why the demonstration. If you don't like it, feel free to generate your own/tweak the prompt. I think removing it completely is a net-negative experience for the reader here. Sohom (talk) 12:36, 25 June 2025 (UTC)

@Sohom Datta: No it was not - the dictionary was added on 15 March having been published on 8 March. The first example (which has the same problem as the later one) was added on 20 March. Your claim that it refers to any AI-generated code is patently incorrect per the quote from Simon Willison. SmartSE (talk) 13:29, 25 June 2025 (UTC)

Fair enough, but I still think a example will be beneficial to folks, removing it wholesale is a disservice to our reader since it makes things less interactive/visual to the reader. Also I will note that Simon Willison has himself pointed out that the mainstream usage of the term has been shifting to refer to any use of AI code [1] (he is against it, but that is his POV on the topic). Sohom (talk) 16:09, 25 June 2025 (UTC)

Providing misleading information to readers is a far bigger disservice. SmartSE (talk) 16:45, 25 June 2025 (UTC)

I don't think it is misleading when a large part of mainstream media mis-characterizes the topic as well. Our articles should reflect both the academic/tech consensus as well as what folks on the ground think. That being said, subtle difference in prompting was a oversight on my end, not a deliberate attempt to mislead the audience. If my goal was to mislead people I wouldn't be spending hours reading through and understanding research papers and community publications to write about technology. Implying that it is a deliberate attempt to mislead folks is uncalled for.

Trying to bring this back to the more constructive side, would [2] be a more appropriate illustration ? Sohom (talk) 17:01, 25 June 2025 (UTC)

File:ExampleOfVibecoding.png Different suggestion from a screenshot of Chat-GPT, we could use a screenshot from an IDE of something being vibe coded. I feel like this is better because this is more in line with how actual programmers vibe code these days, with the LLM integrated into the IDE. Example given (took prompt suggestion from Sohom Datta) is using VS Code, but feasibly we could use another IDE if its more photogenic. //Lollipoplollipoplollipop::talk 17:33, 25 June 2025 (UTC)

Honestly that could be a good idea as well! Sohom (talk) 17:38, 25 June 2025 (UTC)

IMO the prompt get a random Wikipedia page is an example of an initial vibe coding prompt and overcomes the deficiencies of the earlier 'example' (stating a language and even a specific method to use). However, it seems pretty clear that vibe coding means much more than feeding in the initial prompt. After all, the term is vibe coding, not vibe prompting.

-- Jmc (talk) 20:18, 25 June 2025 (UTC)

"Chat-GPT, write me a generic Wikipedia article justifying an inapt neologism."— Preceding unsigned comment added by 2600:1700:6ae5:2510::44 (talk) 17:26, 29 October 2025 (UTC)

Excessive Citations

The lead has quite a few references (4-5) for the same general statement summarizing Vibe Coding. This seems excessive and is a maintenance headache. — Safety Cap ¯\_(ツ)_/¯ (talk) 19:24, 3 June 2025 (UTC)

This now appears to be fixed. I'll remove the banner. AndyGordon (talk) 13:45, 28 June 2025 (UTC)

Tea app

The Tea app is famous Vibe coding case:

https://www.bleepingcomputer.com/news/security/tea-app-leak-worsens-with-second-database-exposing-user-chats/

https://decrypt.co/331961/tea-app-claimed-protect-women-exposes-72000-ids-epic-security-fail 2A02:16A:A201:0:127C:4CFF:FE68:5CD5 (talk) 09:52, 30 July 2025 (UTC)

Purportedly. kencf0618 (talk) 02:14, 4 August 2025 (UTC)

Nomenclature, The Democratization of Code, and DeepSeek.

"Vibe" as per the OED dates to 1968. Unsurprisingly. Keeping WP:NOTADICTIONARY in mind, should the etymology be included? (Good Vibrations, anyone?)

Secondly, this remark by Andrew Chen is on-point: "Most code will be written (generated?) by the time-rich," he said. "Thus, most code will be written by kids/students rather than software engineers. This is the same trend as video, photos, and other social media." (Check history for decrypt.co citation.) The printing press, the Brownie camera, the AK-47, Feynman diagrams, public-key encryption, and the personal computer itself all democratized what was esoteric. Large languages models (LLM's) are doing likewise. It'll be fascinating to watch this play out.

Finally, E.g, I have dyscalculia. I can't program, yet made five Github commits this afternoon, crediting DeepSeek. The street finds its own uses for things, to quote Bruce Sterling. That's fun; that's a hobby. Forensic coding is going to be about what's been shoved into production, to disastrous effect. In this wise pay attention to The Economist and the technical literature. kencf0618 (talk) 02:50, 4 August 2025 (UTC)

alternative approach section

i have doubt about this section

https://en.wikipedia.org/w/index.php?title=Vibe_coding&diff=1310724378&oldid=1310554328

1. first link to external to insightos.app . it is not recommended not put on body of article see wp:EL
2. first reference is actually link to empty page (archive )
3. Platforms such as Insight OS this part link to (. from what i understand insight os is just a demo
4. second part of first paragraph about delphi ai could be valid but have the same problem wp:EL like the first part
5. second paragraph need citation wp:CITENEED

other but maybe related

• searching org360 result other page https://en.wikipedia.org/wiki/360-degree_feedback
• search persona360 (linked on insightos.app and org360.app) dont give any result

about user contribution

https://en.wikipedia.org/wiki/Special:Contributions/Technological_Tiger

1. https://en.wikipedia.org/w/index.php?title=Organizational_culture&diff=1306451030&oldid=1305114456
2. https://en.wikipedia.org/w/index.php?title=Employee_engagement&diff=prev&oldid=1306442814
3. https://en.wikipedia.org/w/index.php?title=360-degree_feedback&diff=prev&oldid=1304507303

these mention org360 as well, so i think this could be case of wp:PROMOTION

(note i still havent fully check 360-degree feedback page)

or maybe im wrong.

can you comment on this @Technological_Tiger

Lokiretro (talk) 11:43, 11 September 2025 (UTC)

Hi, I see what you mean, thanks for the feedback. Let me re-work the article, remove the links, and find other examples. Technological Tiger (talk) 16:01, 11 September 2025 (UTC)

Do we really need a 'Tools and platforms' section?

I'd argue that such a section is fundamentally at odds with the essential concept of vibe coding, which is, as Karpathy has expressed it, letting LLMs generate all code, IOW, by simply and directly prompting a LLM to generate code, without the intermediate intervention of separate tools and platforms. -- JMC (talk) 02:23, 10 October 2025 (UTC)

i would rather keep it but maybe move it somewhere. in current form it is too small for its own section

i also have seen this edit https://en.wikipedia.org/w/index.php?title=Vibe_coding&diff=1316009722&oldid=1314801464

while this edit can expand the section, notability is kinda questionable Lokiretro (talk) 07:08, 10 October 2025 (UTC)

I'm more inclined to remove it because any LLM can be used to vibe code and having sections like this is always an invitation for people to add more and more examples (often poorly sourced or spammy). SmartSE (talk) 11:08, 10 October 2025 (UTC)

Not wholly against it but it does seem to be a magnet for bad edits advertising their favourite tools. //Lollipoplollipoplollipop::talk 12:32, 10 October 2025 (UTC)

I'll remove it. Two crucial comments: (SmartSE) "any LLM can be used to vibe code" and (Lollipoplollipoplollipop) "it does seem to be a magnet for bad edits advertising their favourite tools". -- JMC (talk) 19:38, 10 October 2025 (UTC)

Clarification on sourced addition

I’ve added back a brief, neutral paragraph with citations to reliable secondary sources, making sure it stays in line with WP:NPOV. If there are any thoughts or concerns about tone or sourcing, I’d be glad to work together here to refine it before any further edits. Hoalsikeer (talk) 18:59, 23 October 2025 (UTC)

Vibe coding was a thing before 2025

Vibe coding was definitely a thing in 2024; I tried it for a bit before deciding that it was extremely unprofessional to vibecode stuff. And you don't even need those big, agentic Claude thingies; all you need is Gemini and a lot of patience.

Would someone please find a reliable source confirming this and add it to the article? GrinningIodize (talk) 16:36, 1 December 2025 (UTC)

Proposed expansion: Security concerns

The existing "Concerns" section mentions security vulnerabilities briefly. I propose expanding with specific research findings:

Current text

"Vibe coding has raised concerns about understanding and accountability. Developers may use AI-generated code without fully comprehending its functionality, leading to undetected bugs, errors, or security vulnerabilities."

Proposed expansion

Research quantifies these security risks. A 2023 Snyk survey found that 56.4% of developers reported AI coding tools sometimes or frequently introduce security vulnerabilities, yet 80% of developers admitted to bypassing established security policies when using these tools. Only 10% of developers scan most of the AI-generated code they use.

The practice of accepting code without review—central to vibe coding's definition—amplifies these risks. Simon Willison noted that if code is "reviewed, tested, and understood," it is not vibe coding, implying that true vibe coding inherently bypasses security review processes. Security researchers warn that AI-generated code may contain vulnerabilities stemming from outdated patterns in training data, which developers practicing vibe coding would not detect.

Radius314 (talk) 03:04, 15 December 2025 (UTC) Radius314 (talk) 03:04, 15 December 2025 (UTC)

I appreciate the intention here, but the sources need work. Cybersecurity Dive may (or may not) be reliable, but Cloud Wars, Snyk, and other Wikipedia articles are not inherently reliable and should not be used by themselves. Summarize reliable, independent sources without vagueness, please. Grayfell (talk) 19:54, 28 December 2025 (UTC)

I would consider Snyk reliable, Cloud Wars is up for debate. Sohom (talk) 20:19, 28 December 2025 (UTC)

To be reliable, we need to see a reputation for accuracy and fact checking. Since Snyk is neither academic nor journalistic, the best way to cite Snyk is via reliable, independent sources with editorial oversight. Grayfell (talk) 20:27, 28 December 2025 (UTC)

@Grayfell Snyk is the company which manages most vulnerability disclosure for almost all of the JavaScript ecosystem. If we have a problem with their accuracy and fact checking, we can never trust almost any vulnerability in the JavaScript ecosystem and the global cybersecurity ecosystem needs to be very afraid. (To put things into perspective, what you are doing here is saying, "we should not cite the Securities and Exchange Commission report on bitcoin they do not have a history of being accurate or fact checking" which is just plain absurd.) Sohom (talk) 21:26, 28 December 2025 (UTC)

Did you personally vet these sources? You just cited Wikipedia in your third source ... Caleb Stanford (talk) 21:13, 28 December 2025 (UTC)

Did you use AI to generate the proposed expansion? Caleb Stanford (talk) 21:17, 28 December 2025 (UTC)

References

Linus Torvalds case

Hi @Jmc: maybe we can figure this out here (re diff)

It was reported in January 2026 that Linus Torvalds had made use of Google Antigravity to generate a Python‑based audio sample visualizer.component of AudioNoise, an application focusing on digital audio effects and signal processing. Torvalds explained in the project's README file that, rather than writing all the code himself, "the Python visualizer tool has been basically written by vibe-coding."

3 things - (1) what is visualizer.component? I am not understanding the . formatting. (2) why do we need "an application focusing on digital audio effects and signal processing" ? Sounds redundant with "audio sample visualizer", I don't see what it adds. (3) Where in the source are you seeing that it's a component?

Thanks! Caleb Stanford (talk) 20:05, 19 January 2026 (UTC)

@Caleb Stanford Thanks for taking to Talk. To answer your question 3 (and implicitly answer the others), "While Torvalds hand‑coded the C components, he turned to Antigravity for a Python‑based audio sample visualizer" [as another component of Audionoise].

-- JMC (talk) 20:20, 19 January 2026 (UTC)

That's your own addition and not in the source? Does not address my questions (1) and (2) Thanks, Caleb Stanford (talk) 20:45, 19 January 2026 (UTC)

No, it's a direct quotation from the source. -- JMC (talk) 20:50, 19 January 2026 (UTC)

You added [as another component of Audionoise]. Your text is not a clear summary of the source, and have not answered questions (1) and (2) Caleb Stanford (talk) 20:52, 19 January 2026 (UTC)

To answer your question 1 explicitly (give me a little time and I'll answer your question 2 in detail, and show why I characterised your own edit as an inaccurate condensation): the dot between visualizer and component was simply an unfortunate stray, since removed. -- JMC (talk) 21:19, 19 January 2026 (UTC)

It's instructive to look at the GitHub repo to get a clearer sense of the project than that given by ZDNET's popular account. (I'm aware that I'm delving into a primary source here, but that's purely for the purpose of elucidation and is not intended for use in the article itself.)

The overall function is that of a guitar pedal and is coded in C. The visualiser tool component is separately coded in Python and (to quote from the README) "has been basically written by vibe-coding".

In the ZDNET source, the overall function becomes a "program called AudioNoise -- a recent side project focused on digital audio effects and signal processing" and the visualiser tool component of it becomes "a Python‑based audio sample visualizer".

You conflated the two (the overall C program named AudioNoise and its visualiser tool Python component) in inaccurately condensing them into a Python‑based audio sample visualizer called AudioNoise (to quote from your edit).

-- JMC (talk) 22:22, 19 January 2026 (UTC)

Thanks for the reply. OK about the visualizer.component, but I find it really jarring that after I edited to improve your text, on multiple occasions you copied back in the misformatted text, and didn't seem to acknowledge the mistake. If people are editing your work, they are probably trying to help out, maybe check what they fixed before assuming they messed up your intent? And please be more careful about copyediting.

Heard about the distinction between audionoise etc. ... but is this really that relevant to the article. Can we please summarize this more briefly? We really do not need "an application focusing on digital audio effects and signal processing". It is not clear that we need the mention of Torvalds at all, but I'm not contesting that at the moment. Caleb Stanford (talk) 04:22, 20 January 2026 (UTC)

Thanks for your feedback. Given Linus Torvald's stature, his use of, and opinion about, vibe coding are indeed noteworthy.

The distinction between the encompassing human-coded AudioNoise project and its vibe-coded visualiser tool component is germane. At the same time, I do concur that "an application focusing on digital audio effects and signal processing" could be more succinctly worded. I'll work on a rewording. But I do have to say that I'm very disappointed that you've unilaterally gone ahead and made alterations before we've concluded our collegial discussion here.

-- JMC (talk) 06:54, 20 January 2026 (UTC)

Sure. Any way you can word it -- concisely please and clearly -- is fine with me. Not my intention to edit unilaterally, just to take a stab at clarifying the wording now that I understood from you why you took issue with the previous shortening. Caleb Stanford (talk) 07:22, 20 January 2026 (UTC)