I understand that this is a hot topic, and I’m honestly not trying to start a flame war here. I’m writing a blog post and trying to think through the topic, and thought folks smarter than me that hang out here might have some opinions on the matter.
I can see an argument for being something along the lines of:
Yes, because the product is the product, no matter how it’s made.
Yes, because it doesn’t matter if it’s AI-generated or not as long as it does the job.
Yes, AI is the future of all computing, get with the program.
But I can also see arguments against being:
No, why bother paying someone else to prompt Claude for me? I can do that myself.
No, because of ethical reasons.
No, because of code-quality or security concerns.
No, because I worry about the long-term viability of the project.
What do you think? Do you have any concerns about the current wave of vibe-coded apps hitting the market?
If you use it and there’s.a method to pay for it, pay.
If you’re not paying for something that is fee-based, don’t use it. That’s theft.
My concern with publicly released vibe-coded apps by inexperienced devs are that they won’t be set up for proper testing and QA, documentation, or support.
I think custom designing for yourself is awesome. I think sharing with a few friends who can give you feedback, also super. But releasing it widely, even if it’s free, worries me.
I’ve seen what happens with even small bugs on a publicly released app. Pushback can turn to vitriol so quickly, even if you have support staff answering appropriately, and are transparent about the bug and the fix.
Yes, if they ask for payment. Theft is never justified.
Then do it yourself. It will be better matched to your use case.
You can get too picky about this. Using compilers has the computer writing the machine code for you. So is the Swift written app unethical when the author could have written in machine code?
These are concerns with any app, no matter how it was made.
Not really. No different than other under-developed apps.
The more important, expensive and/or complex the software, the more mindful I am about its provenance. But that applies to more than just whether its vibe coded or not.
Mathematica is very complex, robust software, whose price can run into the thousands of dollars. If Wolfram used AI to generate all of the codebase – which they are more than capable of doing – would it be a no-buy for that reason?
If someone conceived an idea and then hired someone else to write the code, would you pay for that when they didn’t write the code themselves?
I’m a computer engineer by education, but I’ve not a ton with it since graduating 15 years ago. AI has let me start building some things that I’d have no business building without months of learning because I can have the idea, know how to break it down product-wise, but not need to sling the actual code. The newest models are really good with code quality, but I wouldn’t have trusted them as much prior.
My point: the idea and execution thereof (which includes code quality) is what’s valuable, not the way in which the code was written. If it solves a real problem you don’t want to build yourself, why not pay for it?
(BTW, if you’ve seen the codebases for legacy apps you use for pretty critical functions like banking, you’d be appalled at some of the code written by humans…)
A Claude Skill is an “app” every bit as much as an app written in assembly or C or Python. Sure parts of it are non-deterministic - that is a feature of the “English” coding language just like C or Pascal have their quirks. Plus a Claude Skill can have deterministic parts written in a classic coding language.
Saying Claude skills are not “real coding” is like saying in 1965 that compiled languages are not “real coding” since they are not in assembly language.
It’s just a new abstraction layer with different syntax.
Even early compiled and interpretative languages had non-deterministic capabilities such as a random number generator or If statements tied to a system clock.
AI isn’t alive and does not think - it’s just an extremely sophisticated run-time computer language.
A Claude “Skill” is a markdown file containing plain English instructions and maybe some example code to feed to the LLM.
An “Agent” is little more than a shell script reading input from either files or stdin, sending that input to the LLM for interpretation, and executing the commands it receives back.
A compiler doesn’t need an internet connection, and it doesn’t require a datacenter with the energy requirements of a small town to run. Turn off the internet, and python3 still works, so does clang or gcc.
The robotics used in automotive manufacturing is also not the same thing. You don’t get 100 different kinds of cars out of a single factory assembly line.
If you can do it yourself, then do it yourself. But not all “vibe coded” apps will be something you can do yourself. For starters, did you even think of the idea before you saw the app?
I can assure you that LLM involvement is absolutely not a requirement for this! Does it make this more likely? Probably, but there have been badly documented (or just un-documented) and buggy apps that get little to no support for a long, long time.
Then there’s the question of… if you didn’t know it was “vibe coded”, would you buy it? How much do you know about the developers of apps in general? Probably very little in many cases.
I have four apps in the App Stores currently. Two of them were entirely "vibe coded”. I always put that phrase in quotes because I think it has taken on a pejorative meaning. My apps were developed with the aid of a LLM. I put a lot of effort into those two apps.
The other two were written “by hand” by me some time ago. I can assure you that when it comes to iOS development (which these two are), I am not a very good developer. Would you prefer to buy those from an inexperienced dev who rarely does updates? What if I told you both of those apps recently received updates, and the primary reason why is that I used LLM support to do so. I didn’t just add features, I had the LLM update the internals to more modern technologies and generally improved the “shonky bits”. Now would you buy them?
I think so, so much discussion of “AI”/“LLM” use is overly reductive. Would you buy a house from someone who has never used a hammer, only a nail gun? I suspect you’d want to know a lot more about how the house was built.
Dave Rogers, who stops by MPU Talk from time to time, made a compelling argument about the ethics of selling an app that’s 100% LLM-generated, or “vibe coded”:
Where I do have some objection is when someone uses an AI product to create something that is totally outside the scope of what they might otherwise have been able to achieve without it, and then seeking to monetize that creation.
…
But I don’t think one needs to wait years to make a personal choice based on one’s own moral center. If you’re seeking to profit from others’ labors for little more than the simple act of writing a good prompt, I think that’s wrong.
His argument has bothered me, in a good way, ever since I read it.
There’s something inherently exploitative about LLMs that rubs me the wrong way. Even as an experienced software developer, I’m still puzzling through the ethics of all this. Thank you all for sharing your thoughts! This is a topic worthy of serious discussion.