We Should Standardize Vibe-Coding Ethics

Almost anyone reading this probably has thought the following statement:

“There’s so much cool software coming out that it’s hard to even keep up. Most of it is slop.”

I’ve covered this on my channel plenty.
AI music solves the imaginary problem of there not being enough music on Spotify.
AI image generators solves the imaginary problem of there not being enough images on the internet.
AI video makers turn 30 minutes of keyframing into hours of expensive prompt-gambling.

The loudest part of the AI bubble is billions of dollars competitively making services that attempt to do a person’s hobby or dream job. Don’t get me wrong, the novelty of quickly making a pixel art version of a scene I described hasn’t completely worn off. But it’s also not exactly accomplishing much beyond entertainment for most people.

Vibe-coding, on the other hand, immediately solves a lot of problems.
For me, it’s finding a quirky repo on Github and letting Hermes deal with sorting through the outdated libraries on a VPS. It’s turning my Pure Data or SuperCollider sequencers into packages I can share as audio plugins. It’s preparing 5GB of CSV data for visualization.

I hated the idea of vibe-coding 5 years ago. I was agnostic 2 years ago. Now, it is unquestionably saving me time, and if I’m particularly careful, it is doing so without compromising quality. The ad-nauseum marketing didn’t work on me. I started using coding agents as a last resort, and now I’m learning how to improve their output rather than learning specific library definitions that I’d never use again.

Of course I’ll be criticized for this take. There’s an entire community of people who bully people for using AI anything, most tragically when someone posts their own artwork or music. This conversation isn’t for them. I believe that their etiquette has harmed independent artists and creators more than AI has.

The Nuance of Easy

Let’s imagine a realistic scenario:

Joe has developed music software professionally for 30 years. If a synthesizer filter sounds too weak, Joe can sort through the DSP code and figure out exactly why, then tune it to his experienced liking.

Jason is a seasoned musician who has a lot of legitimately good ideas for music software but has struggled with learning the syntax to develop them. Now, Jason can make a filter for a synthesizer by using an LLM that may have been trained on some of Joe’s older code.

As a musician, at this juncture of my example, I’m extremely grateful for both Joe and Jason because they may inspire me and make my hobby more enjoyable. Win/win.

Enter Capitalism

The problem is that Jason’s software may end up being sold in the same market as Joe’s.
Joe’s career is not at risk because AI agents are taking his job, but that there are 5x more people in Joe’s slim industry-demographic selling and marketing software that claims to do the same thing as Joe’s software. Since it was vibe-coded, it can be priced much more competitively. But Jason’s software almost certainly won’t sound as good, run as efficiently, or be as stable as Joe’s software, but now the burden is on Joe to advertise this. Quality takes a back seat to volume.

What’s worse is that Jason is incentivised to lie. Can Joe prove that Jason used AI? How can Joe make a feature that Jason’s AI agent can’t emulate? This example is starting to sound like runaway fiction, but I guarantee you that this exact scenario is playing out in nearly every specialized software industry.
Ultimately, the victim in this example is the person shopping for some software to make music.

What exactly is AI accomplishing here? A lot.
One can argue that more software democratizes and humbles price-gougers. The antagonist and protagonist in my example would change sides pretty quickly if I told you that Joe was Adobe. Most of us have opinions on AI that are contingent on the characters being described, not the technology itself. That’s a good thing, because no matter what Sam Altman wants you to believe, an LLM is not capable of being good or evil, but just a tool used by good or evil actors.

A Call To Formalize AI Cancel Culture

A few months ago, I decided to buy a digital drawing tablet and with my own lack of expertise, open up Adobe Illustrator and make my own video illustrations. I did this knowing for certain that I would be attacked in the YouTube comments for using AI, as I had mentioned above, any artwork being shown to the public in 2026 is treated with vitriol and skepticism. But the irony is that I did use AI, and a whole lot of it. Adobe’s AI corrected wiggles in my curves by using machine learning models trained to interpret input to produce cleaner paths. I brought in some sketches from Procreate, and AI traced it into vector graphics.

Ultimately, the big difference between me drawing into a heavily assisted interface or typing out prompts was that drawing was the option that was enjoyable to me. So that begs to question, why exactly would I have a problem with someone using an image generator if typing out prompts is enjoyable to them? Because I looked more like a “real artist” while doing it? Because I used electricity and GPU power differently?

The point I’m trying to arrive to is that this entire battle is extremely abstract and nuanced unless capitalism enters the chat. A better way to police AI ethics would not be by focusing on the artwork, music, or software itself, but the transparency and potential problems that arise when it enters the real world.

Since someone releasing AI music or putting AI generated images in their videos has little tangible effect on anyone, this stream of thought is focused on software. You can preview a song or look at an image, but you need to install software. In many cases, you need to pony up and buy it before you can even verify that it runs correctly, and it’s rarely refundable.

Proposed Rules For Ethical Vibe-Coding

A. Open source the code.

This is the hardest sell, but I think it’s the most important. Coding agents are quite literally built on the labor and generosity of open source developers, and the licensing terms and attributions are lost in the process. LLMs make it all of a sudden legal to sell something with a non-commercial license or without citing the developer.

To state the obvious, if you vibe-code something, you don’t know how it works. You didn’t read the Github issue explaining why it won’t work with the most recent file compression library. You may not even know that your software calls a file compression library. So when you vibe-code a fix for a GUI bug and the entire program breaks, you may get stuck in vibe-code hell trying to figure out why, and your customers will never be able to get an update for something they paid for.

Publicly updating a repo helps with this, a lot. You can ask a friend, hire a developer, or even use another LLM to comb through your nightly changes and issues to figure out exactly what went wrong. Without this forensic ledger, you might be forced to start over from scratch.

Open sourcing your code doesn’t mean you can’t sell the software. Plenty of commercial software suites simply keep the code open while selling the builds or installers. I think this is currently the only way I can ethically justify selling vibe-coded software, because otherwise you’re effectively selling software you cannot confidentially support.

B. Transparency by Grading

In the software defined radio world, vibe-coding is running rampant. The good news is that due to the open culture of the hobby, the vast majority of SDR software is open source. The bad news is that AI-assisted software often results in wonderful, friendly, and innovative user interfaces that lack the expertise required to synchronize a clock and demodulate RF from proprietary hardware and convert it to nicely dithered and noise-corrected audio. In English, this means that you have really accessible radio interfaces that can barely tune into radio signals. The chef’s kiss of good SDR software is in the DSP, and most experienced users will subject themselves to exhaustive amounts of UI pain if it means they can get accurate results from the hardware and antenna. AI may one day solve this. That day is not today. In the meantime, amidst the AI flame-wars happening on a ham radio forum, I proposed the following:

Developers should obviously practice transparency about AI use, but as I mentioned in my Illustrator example, it’s pretty granular and nuanced. What if an ethical self-rating system was standardized for developers?

NoAI
No AI of any type was used in the development of the software.

AI-aux
AI was only used in an auxiliary manner. For example, an AI agent that updates binaries or automates your workflow, but no AI-written code is present in the software.

AI-UI
AI may have been used in an auxiliary manner, but AI-written code is limited to the user interface or graphics. If you were to run the software’s operation from a terminal, no form of AI-written code would be used.

AI-dev-A
AI-written code is used throughout the software, but is thoroughly verified, tested, and understood by an experienced developer who could have otherwise written it themselves.

AI-dev-B
AI-written code is used throughout the software and has assisted in areas that an experienced developer could not verify or completely understand.

AI-dev-C
AI-written code is used primarily throughout the software under the guidance and supervision of a novice developer.

AI-V
The software’s code is entirely written by AI, and currently support is contingent on AI.

I’m sure there’s a lot I’m missing, but I feel like this would prevent a lot of social and ethical problems in this area. A lot of problems right now come from the lack of a declaration formality when sharing software or listing it for sale, and it seems to only benefit bad actors. Developers using AI in good faith are encouraged to not disclose its use due to the vague stigma, and then users feel betrayed upon suspicion of vibe-coding. An expected rating would not only alert users to software with a potential for lower quality, but it would devilify AI’s use when this disclosure is applied to software with a solid reputation.

Of course, you’ll have developers that misrepresent their rating, but at least there would be a standard beyond looking at a screenshot and deciding whether or not you think the software was vibe-coded, and thus worth your time or money.