Thank you @andrewchen, it's been awesome to prompt that game into existence. I hope you don't mind if I share links? You can try the game on fly.zullo.fun ;) or follow me x.com/NicolasZu if you want more insights on how I did it.
Very interesting take - I also think vibe coding will create a new category of creators that build products through conversation, and has also become the easiest on-ramp to code for non-coders.
In many cases, the "product" will be for just one user, and often even for only one use. It's already happened to me multiple times.
Generative AI may be enabling a sort of Moore's Law for software development, however we still live in a world full of legal and business complexities that aren't going away anytime soon. Publishing a commercially viable app requires (usually) things like a Terms of Service, Privacy Policy, contracts with third parties for APIs to do actual stuff like sending an email, SMS or accessing data, etc. I do use GPT for creating code, however I've been a developer for 30 years, so I know what to ask and I know how to review. GPT will gleefully produce code that encodes an email address as a GET URL query parameter (illegal in most States) or assigns sequential user ids instead of using a hash (easily subject to hacking). Can these tools get better? Maybe. Understanding the nuances of compliance, marketing, business services and security goes far beyond "create me a login page using nextjs" and getting some TSX file out. Will it know to hash your password with PBKDF2 or simply choose SHA256? As an app designer (and not a developer) will you understand the implication? My take is that these tools are super valuable for veteran programmers who don't want to be bothered with mundane tasks and have the experience to spot issues, but they are still inappropriate for junior developers, or worse, people with no coding experience at all. Can enough "parameters" be added to make it work? We are already seeing the logarithmic flattening of parameter counts in models over time. Maybe some training breakthrough will come that allows quadrillion parameter models. In the meantime I think non-developers are relegated to the realm of minecraft flightsim, which by the way, could have been hand coded in 10x less time ;)
GPT might be your enthusiastic intern, but it’s still the kind that would confidently send out your production credentials in plaintext because “it looked tidy.” We’re entering an era where non-devs can vibe code their way into legal liability at scale. Yes, it’s Moore’s Law for code, but it's Murphy’s Law for compliance. Until LLMs start offering to hash your passwords and walk you through GDPR, they’re best used by devs who already know the difference between a login form and a subpoena.
I agree with everything you said about vibe coding and the predictions, but seriously, where are you seeing a 5:1:1 ratio of engineers to designers to PMs?! From my experience, this is not close to the ratio in Big Tech and PE porticos. It's more like 30:2:1.
I’d say a ratio 30:2:1 comes from the sheer complexity of serving millions of users and well-paying customers. My question about “vibe coding” — can it do that? Surprisingly few people make an income off YouTube compared to the volume of content generated. I assume vibe coding cannot improve on the economics
"But it doesn’t matter, simply because the sheer quantity of content — that is comprehensively able to fill every niche and nuance of demand, and created by real people as the creators (not corporations), make social media dominant. The same could be said of software."
I disagree. Mainly because software and content are two different things. Software in a lot of ways is utility which a person proactively uses. Whereas content is something a person passively consumes and is usually ephemeral. I believe it's been a proven fact that with software quality wins...
We already have this. It's called shovelware, and it drowns out valuable and useful software by orders of magnitude on every platform. As more and more people do this, more and more software will have the same bugs, the same security flaws, look the same and feel the same. For all the quantity of content on Youtube, the SUCCESSFUL content has fallen back to needing video editing, lighting, and other production skills. Exceptions exist, of course, but the existence of successful reality TV shows didn't supplant well-written TV shows.
Welcome to the era of everyone making a new game or software. They'll all be terrible and you'll find it as challenging to find a good one as you do finding valuable results on your average social media site.
Very interesting punditry, which I largely agree with. Especially the part about vibe coding graphically, which is something we’ve introduced into our no-code platform. AI aided coding will remain important for building some components, but the conventional code paradigm is not a good format for either machines or humans in dealing with millions of permutations that arise when building for real (as you also allude to)
I’m extremely bullish on distribution, advertising, collectives, and communities. Creators that are money-rich and time-poor and going to increasingly rely on paying their way to the top. We’re also going to see more groups of designers, writers, and engineers pull their audiences together in order to cut through the noise.
The discussion on hiring is interesting. I like their points of assessing "taste" and "systems thinking." Taste can involve debugging and seeing how they review code. The system design interview can remain.
I completely agree about the quality of the generated code improving.
Cursor (and others) are gathering now a goldmine of data - from every error, and every fix we ask, they’ll learn. I’m pretty sure the limit will be our own ability to express ourselves clearly…
“‘The code grows beyond my usual comprehension’ might be the most honest line in the whole AI dev discourse right now.”
Loved this. What stood out most to me is how you framed vibe coding not as a gimmick, but as a cultural tipping point — a kind of Cambrian explosion for software, where the constraints are no longer technical. They’re creative, distributive, and motivational.
This is an excellent article! I put together https://www.whatisvibecoding.com/ to try and help others understand what Vibe Coding is and how to start it.
Thank you @andrewchen, it's been awesome to prompt that game into existence. I hope you don't mind if I share links? You can try the game on fly.zullo.fun ;) or follow me x.com/NicolasZu if you want more insights on how I did it.
appreciate your posts and everything you're sharing! Your work def inspired this essay.
Very interesting take - I also think vibe coding will create a new category of creators that build products through conversation, and has also become the easiest on-ramp to code for non-coders.
In many cases, the "product" will be for just one user, and often even for only one use. It's already happened to me multiple times.
I think it has already become a superpower people can add to their "talent stack", accessible to anyone. After using it for a while, I ended up introducing the skill to friends and colleagues, and even created a one-hour crash course to prove it can be done real fast (Udemy: https://www.udemy.com/course/vibe-coding-with-chatgpt-and-python/?referralCode=BD3711D8DA286AD9062A).
Vibe coding with neuralink will be wild
Generative AI may be enabling a sort of Moore's Law for software development, however we still live in a world full of legal and business complexities that aren't going away anytime soon. Publishing a commercially viable app requires (usually) things like a Terms of Service, Privacy Policy, contracts with third parties for APIs to do actual stuff like sending an email, SMS or accessing data, etc. I do use GPT for creating code, however I've been a developer for 30 years, so I know what to ask and I know how to review. GPT will gleefully produce code that encodes an email address as a GET URL query parameter (illegal in most States) or assigns sequential user ids instead of using a hash (easily subject to hacking). Can these tools get better? Maybe. Understanding the nuances of compliance, marketing, business services and security goes far beyond "create me a login page using nextjs" and getting some TSX file out. Will it know to hash your password with PBKDF2 or simply choose SHA256? As an app designer (and not a developer) will you understand the implication? My take is that these tools are super valuable for veteran programmers who don't want to be bothered with mundane tasks and have the experience to spot issues, but they are still inappropriate for junior developers, or worse, people with no coding experience at all. Can enough "parameters" be added to make it work? We are already seeing the logarithmic flattening of parameter counts in models over time. Maybe some training breakthrough will come that allows quadrillion parameter models. In the meantime I think non-developers are relegated to the realm of minecraft flightsim, which by the way, could have been hand coded in 10x less time ;)
GPT might be your enthusiastic intern, but it’s still the kind that would confidently send out your production credentials in plaintext because “it looked tidy.” We’re entering an era where non-devs can vibe code their way into legal liability at scale. Yes, it’s Moore’s Law for code, but it's Murphy’s Law for compliance. Until LLMs start offering to hash your passwords and walk you through GDPR, they’re best used by devs who already know the difference between a login form and a subpoena.
Awkward
I agree with everything you said about vibe coding and the predictions, but seriously, where are you seeing a 5:1:1 ratio of engineers to designers to PMs?! From my experience, this is not close to the ratio in Big Tech and PE porticos. It's more like 30:2:1.
I’d say a ratio 30:2:1 comes from the sheer complexity of serving millions of users and well-paying customers. My question about “vibe coding” — can it do that? Surprisingly few people make an income off YouTube compared to the volume of content generated. I assume vibe coding cannot improve on the economics
"But it doesn’t matter, simply because the sheer quantity of content — that is comprehensively able to fill every niche and nuance of demand, and created by real people as the creators (not corporations), make social media dominant. The same could be said of software."
I disagree. Mainly because software and content are two different things. Software in a lot of ways is utility which a person proactively uses. Whereas content is something a person passively consumes and is usually ephemeral. I believe it's been a proven fact that with software quality wins...
We already have this. It's called shovelware, and it drowns out valuable and useful software by orders of magnitude on every platform. As more and more people do this, more and more software will have the same bugs, the same security flaws, look the same and feel the same. For all the quantity of content on Youtube, the SUCCESSFUL content has fallen back to needing video editing, lighting, and other production skills. Exceptions exist, of course, but the existence of successful reality TV shows didn't supplant well-written TV shows.
Welcome to the era of everyone making a new game or software. They'll all be terrible and you'll find it as challenging to find a good one as you do finding valuable results on your average social media site.
On point. Also just wrote about vibe coding a helpful assistant yesterday https://hibag.substack.com/p/chatbot-in-a-trench-coat
Very interesting punditry, which I largely agree with. Especially the part about vibe coding graphically, which is something we’ve introduced into our no-code platform. AI aided coding will remain important for building some components, but the conventional code paradigm is not a good format for either machines or humans in dealing with millions of permutations that arise when building for real (as you also allude to)
I’m extremely bullish on distribution, advertising, collectives, and communities. Creators that are money-rich and time-poor and going to increasingly rely on paying their way to the top. We’re also going to see more groups of designers, writers, and engineers pull their audiences together in order to cut through the noise.
I listened to the Lightcone Podcast episode "Vibe Coding Is The Future". (Y Combinator)
https://youtu.be/IACHfKmZMr8?feature=shared
The discussion on hiring is interesting. I like their points of assessing "taste" and "systems thinking." Taste can involve debugging and seeing how they review code. The system design interview can remain.
Any predictions for hiring processes?
Great article.
I completely agree about the quality of the generated code improving.
Cursor (and others) are gathering now a goldmine of data - from every error, and every fix we ask, they’ll learn. I’m pretty sure the limit will be our own ability to express ourselves clearly…
“‘The code grows beyond my usual comprehension’ might be the most honest line in the whole AI dev discourse right now.”
Loved this. What stood out most to me is how you framed vibe coding not as a gimmick, but as a cultural tipping point — a kind of Cambrian explosion for software, where the constraints are no longer technical. They’re creative, distributive, and motivational.
Thank you
This is an excellent article! I put together https://www.whatisvibecoding.com/ to try and help others understand what Vibe Coding is and how to start it.
It'll be interesting to see the first bank develop its online user system this way...