Wild how many people are posting about "vibe coding" entire apps with AI and claiming they're pulling in actual revenue from production-ready products. Like... really? 🤔
Maybe I'm just old school, but something feels off about going from zero to profitable app in a weekend with mostly AI-generated code. I still choose AI-Assisted coding over vibe coding.
Anyone else a bit skeptical, or am I missing something here?
True. But then there are so many posts and YT videos and reddit posts about it. Shorter timelines (2 hours to 2 weeks) and higher MRR (2k to 20k) and it makes me doubt myself, what am I doing, How come these people, especially those youngsters are doing it so well. I hesitate to lie even in the comment on any post, so its hard to believe for me that people will lie so much in their YT videos and podcasts. Its still a puzzle for me, how can people claim to make money so easily. 10k MRR is like a norm on many YT channels.
They are all fragile and silly/pathetic toy apps, which will break under stringent testing. And yes, many lie about their achievements, and many(folks who know nothing about serious dev), out of ignorance, are so easily impressed, with their new shiny toy app, they are quick to write a blogpost about it. But then, when it blows up in their faces, you will not hear from them.
Here is my RevenueCat showing that I got a paying user https://imgur.com/a/3Ggznoa (wasnt me, I swear). Yes of course it's a measly $4, however, it was just a fun experiment for me to see if I could do it. Especially because the idea for the app (system design flashcards native app) was something my wife asked for when studying for a FAANG interview and I was shocked there was nothing on the app store.
It was not my first app on the app store so I had experience with the process however I did have bolt and then Cursor (with Claude in both cases) write 100% of the code.
I don't understand the skepticism. Obviously you can only make something simple and minimal, but guess what? If you want to quickly monetize an app, you should pick something simple and minimal even if you're writing all the code.
Also, as part of the hackathon I forced myself to do all vibe coding and I learned that many times I would get stuck and the AI would go in circles, it was possible for me to get unstuck with clever prompting, whereas previously Id drop into the code myself. One thing people may be missing is that "prompt engineering" is a VERY real thing, not an exaggerated fake job title.
It's fashionable to bash on AI coding for a variety of reasons (fear, typical engineer pessimist groupthink), but if you don't believe it's possible you could just try to do the thing and find out that it is in fact possible. Or, you could assume everyone's inventing up elaborate lies as part of some very complex grift.
(One last aside, I did use bolt as part of their hackathon so promoted them a bit in a hope to win a prize but having done more vibe coding since, if you're a dev I'd just recommend Claude Code or Cursor as the vibe coding 'wrappers' like bolt and lovable barely add value if you're familiar with creating a dev environment yourself).
I say it like “don’t let AI do TDD, you do TDD with AI” tell it specifically what tests to write when, be in the driver seat and in the navigator seat AI is the plane not the copilot
While I agree that an inexperienced engineer couldn't build an app to production purely with AI, I'll note two things:
* Most of the huge AI failures are mistakes humans make all the time. Your leading example was a query that worked in testing and failed in production. It's absurd to assume that a human engineer would _always_ catch that in review. A sharp human, sure, but many humans would nitpick the syntax then completely miss the huge issue
* While I don't think a human letting an AI do all the work could make it to production, a human that used AI to learn about what it takes to get to production, debugged issues that arose with the help of AI, and generally used it as a partner rather than a crutch could get to production, and much sooner than he previously would.
Let's be honest, it's simultaneously true that vibe coding as the new and better no-code got overhyped, _and_ grumpy, cranky, and scared engineers are downplaying its potential out of a combination of fear and the usual "top comment of Hacker News / Slashdot is always excessively pessimistic" syndrome.
One thing that's certainly true is that nobody even knows what 'vibe coding' even means. The coined was termed to refer to something be done by one of the best engineers in the world yet quickly came to mean something done by non-engineers? AI-Assisted Development is a clearer term but a bit longer to type out...
As a complete aside on code review, one thing I hear teams doing at a YC meetup was reviewing the specs/prompts they feed to Claude rather than the code itself. That way you still get some benefit of review without trying to keep up with an unreasonable amount of code volume. (you allude to this in post just agreeing I also see this trend)
Wild how many people are posting about "vibe coding" entire apps with AI and claiming they're pulling in actual revenue from production-ready products. Like... really? 🤔
Maybe I'm just old school, but something feels off about going from zero to profitable app in a weekend with mostly AI-generated code. I still choose AI-Assisted coding over vibe coding.
Anyone else a bit skeptical, or am I missing something here?
True. But then there are so many posts and YT videos and reddit posts about it. Shorter timelines (2 hours to 2 weeks) and higher MRR (2k to 20k) and it makes me doubt myself, what am I doing, How come these people, especially those youngsters are doing it so well. I hesitate to lie even in the comment on any post, so its hard to believe for me that people will lie so much in their YT videos and podcasts. Its still a puzzle for me, how can people claim to make money so easily. 10k MRR is like a norm on many YT channels.
They are all fragile and silly/pathetic toy apps, which will break under stringent testing. And yes, many lie about their achievements, and many(folks who know nothing about serious dev), out of ignorance, are so easily impressed, with their new shiny toy app, they are quick to write a blogpost about it. But then, when it blows up in their faces, you will not hear from them.
Yeah, totally agree. Test it and you’ll see exposed passwords, personal data and so on.
I literally did it as part of the bolt.new hackathon and documented it on my newsletter:
https://www.aiengineering.report/p/from-zero-to-monetized-ios-app-in
Here is my RevenueCat showing that I got a paying user https://imgur.com/a/3Ggznoa (wasnt me, I swear). Yes of course it's a measly $4, however, it was just a fun experiment for me to see if I could do it. Especially because the idea for the app (system design flashcards native app) was something my wife asked for when studying for a FAANG interview and I was shocked there was nothing on the app store.
It was not my first app on the app store so I had experience with the process however I did have bolt and then Cursor (with Claude in both cases) write 100% of the code.
I don't understand the skepticism. Obviously you can only make something simple and minimal, but guess what? If you want to quickly monetize an app, you should pick something simple and minimal even if you're writing all the code.
Also, as part of the hackathon I forced myself to do all vibe coding and I learned that many times I would get stuck and the AI would go in circles, it was possible for me to get unstuck with clever prompting, whereas previously Id drop into the code myself. One thing people may be missing is that "prompt engineering" is a VERY real thing, not an exaggerated fake job title.
It's fashionable to bash on AI coding for a variety of reasons (fear, typical engineer pessimist groupthink), but if you don't believe it's possible you could just try to do the thing and find out that it is in fact possible. Or, you could assume everyone's inventing up elaborate lies as part of some very complex grift.
(One last aside, I did use bolt as part of their hackathon so promoted them a bit in a hope to win a prize but having done more vibe coding since, if you're a dev I'd just recommend Claude Code or Cursor as the vibe coding 'wrappers' like bolt and lovable barely add value if you're familiar with creating a dev environment yourself).
Interesting article. I don't see how vibe coding is "ideal for learning" though. Really seems the opposite.
Italians are the Kings of airy hand movements, but that doesn't mean that we should be judging their personal life choices.
Really enjoyed this. I actually stumbled on it after seeing it mentioned on Karo’s site. Glad I clicked through, solid read.
I say it like “don’t let AI do TDD, you do TDD with AI” tell it specifically what tests to write when, be in the driver seat and in the navigator seat AI is the plane not the copilot
While I agree that an inexperienced engineer couldn't build an app to production purely with AI, I'll note two things:
* Most of the huge AI failures are mistakes humans make all the time. Your leading example was a query that worked in testing and failed in production. It's absurd to assume that a human engineer would _always_ catch that in review. A sharp human, sure, but many humans would nitpick the syntax then completely miss the huge issue
* While I don't think a human letting an AI do all the work could make it to production, a human that used AI to learn about what it takes to get to production, debugged issues that arose with the help of AI, and generally used it as a partner rather than a crutch could get to production, and much sooner than he previously would.
Let's be honest, it's simultaneously true that vibe coding as the new and better no-code got overhyped, _and_ grumpy, cranky, and scared engineers are downplaying its potential out of a combination of fear and the usual "top comment of Hacker News / Slashdot is always excessively pessimistic" syndrome.
One thing that's certainly true is that nobody even knows what 'vibe coding' even means. The coined was termed to refer to something be done by one of the best engineers in the world yet quickly came to mean something done by non-engineers? AI-Assisted Development is a clearer term but a bit longer to type out...
As a complete aside on code review, one thing I hear teams doing at a YC meetup was reviewing the specs/prompts they feed to Claude rather than the code itself. That way you still get some benefit of review without trying to keep up with an unreasonable amount of code volume. (you allude to this in post just agreeing I also see this trend)