> And then there’s the code that we rely on for bank transactions, package deliveries, medical results, satellite launches, airline flight paths, self-driving cars, mortgage payments, and nuclear power plants. This is durable code, and it’s going to stay that way.
I can tell you from first hand experience that, since long before LLMs were invented, critical software supporting these industries is held together with duct tape and baling wire (and Excel). "Durable" does not mean "good". In my experience production code is often ugly, poorly abstracted, full of special cases and hacks, but most importantly it works.
AI and vibe coding does not prevent the creation of good, robust and durable code. All it takes is for the coder to think carefully about the functions and not fall for the temptation to make the LLM add a bunch of fluff and features "just because they can".
I agree but the problem is that the average developer uses an LLM is to avoid doing so. I know we’re all carefully examining the LLM output here on HN but that’s not how I see a lot of developers work in practice.
I think the big assumption that this article hinges on is that AI will never create durable code. That cannot be true. Now that we’ve finally got a taste of the singularity, humans will never stop trying to bring about AGI, which by definition must write code that is exponentially more durable than anything a human could write. The question is ‘when’.
> The question is ‘when’.
Why ?
In the 60s we thought we'd be living on the moon and mars "soon", flying cars were talked about seriously, that faster than sound jetliners were just around the corner, &c. Sometimes what's possible isn't desirable, or what's desirable isn't possible. No one went back on the moon, space programs are basically dead, planes are still as slow as they were in the 50s, a 747-8 from 2008 isn't much faster than a 707 from 1956
You've wandered away from logic into the realm of wildest fallacy. Your proof about AGI is just a repurposed proof of the existence of a supreme deity! In other words your argument is, "AGI will be the literal most amazing thing in the universe because that's what I define the word AGI to mean." No implementation of AGI could ever satisfy your absurdist threshold because what you want already exists. Code more durable than any written by humans is called DNA. Which DNA and which code persist throughout history is the subject of romances and wars, in other words the stuff of life and death. Only by being wilfully blind to literally everything in the world could you be so ignorant as to believe that you've found some great blank empty slate where nothing yet exists. Sorry, but this space is inhabited already.
Just because you dream of something does not mean it will come true
This library (including the schema documentation) was largely written with the help of Claude, the AI model by Anthropic. Claude's output was thoroughly reviewed by Cloudflare engineers with careful attention paid to security and compliance with standards. Many improvements were made on the initial output, mostly again by prompting Claude (and reviewing the results). Check out the commit history to see how Claude was prompted and what code it produced.
"NOOOOOOOO!!!! You can't just use an LLM to write an auth library!"
"haha gpus go brrr"
In all seriousness, two months ago (January 2025), I (@kentonv) would have agreed. I was an AI skeptic. I thought LLMs were glorified Markov chain generators that didn't actually understand code and couldn't produce anything novel. I started this project on a lark, fully expecting the AI to produce terrible code for me to laugh at. And then, uh... the code actually looked pretty good. Not perfect, but I just told the AI to fix things, and it did. I was shocked.
To emphasize, this is not "vibe coded". Every line was thoroughly reviewed and cross-referenced with relevant RFCs, by security experts with previous experience with those RFCs. I was trying to validate my skepticism. I ended up proving myself wrong.
Again, please check out the commit history -- especially early commits -- to understand how this went.
See the many comments on HN about this
It took an expert and a whole team of software engineer to code "with IA" something well defined, well known, and with already lots of implementation in other languages
Glorious indeed, we have created a new kind of printer ! (irony inside)
(in conclusion, the above post from cloudflare is an ad)
> TO EMPHASIZE THIS WAS NOT VIBE CODED
an expert using high powered autocomplete to successfully take a shortcut towards a well defined problem is not evidence of "AGI" or "the singularity"
belief in "the singularity" is a religious belief. you people sound like evangelicals talking about the goddamned rapture
Sorry, yes, talk of “AGI” or “the singularity” is silly and I should’ve distanced myself from that bit of the post. But the use of AI to write durable code is happening now, like it or not.
It can only be categorized as durable after careful revision from human experts and extensive testing. This is what the article is about.
> And then there’s the code that we rely on for bank transactions, package deliveries, medical results, satellite launches, airline flight paths, self-driving cars, mortgage payments, and nuclear power plants. This is durable code, and it’s going to stay that way.
I can tell you from first hand experience that, since long before LLMs were invented, critical software supporting these industries is held together with duct tape and baling wire (and Excel). "Durable" does not mean "good". In my experience production code is often ugly, poorly abstracted, full of special cases and hacks, but most importantly it works.
AI and vibe coding does not prevent the creation of good, robust and durable code. All it takes is for the coder to think carefully about the functions and not fall for the temptation to make the LLM add a bunch of fluff and features "just because they can".
I agree but the problem is that the average developer uses an LLM is to avoid doing so. I know we’re all carefully examining the LLM output here on HN but that’s not how I see a lot of developers work in practice.
I think the big assumption that this article hinges on is that AI will never create durable code. That cannot be true. Now that we’ve finally got a taste of the singularity, humans will never stop trying to bring about AGI, which by definition must write code that is exponentially more durable than anything a human could write. The question is ‘when’.
> The question is ‘when’.
Why ?
In the 60s we thought we'd be living on the moon and mars "soon", flying cars were talked about seriously, that faster than sound jetliners were just around the corner, &c. Sometimes what's possible isn't desirable, or what's desirable isn't possible. No one went back on the moon, space programs are basically dead, planes are still as slow as they were in the 50s, a 747-8 from 2008 isn't much faster than a 707 from 1956
You've wandered away from logic into the realm of wildest fallacy. Your proof about AGI is just a repurposed proof of the existence of a supreme deity! In other words your argument is, "AGI will be the literal most amazing thing in the universe because that's what I define the word AGI to mean." No implementation of AGI could ever satisfy your absurdist threshold because what you want already exists. Code more durable than any written by humans is called DNA. Which DNA and which code persist throughout history is the subject of romances and wars, in other words the stuff of life and death. Only by being wilfully blind to literally everything in the world could you be so ignorant as to believe that you've found some great blank empty slate where nothing yet exists. Sorry, but this space is inhabited already.
Just because you dream of something does not mean it will come true
Never stop trying != succeed
https://github.com/cloudflare/workers-oauth-provider
This library (including the schema documentation) was largely written with the help of Claude, the AI model by Anthropic. Claude's output was thoroughly reviewed by Cloudflare engineers with careful attention paid to security and compliance with standards. Many improvements were made on the initial output, mostly again by prompting Claude (and reviewing the results). Check out the commit history to see how Claude was prompted and what code it produced.
"NOOOOOOOO!!!! You can't just use an LLM to write an auth library!"
"haha gpus go brrr"
In all seriousness, two months ago (January 2025), I (@kentonv) would have agreed. I was an AI skeptic. I thought LLMs were glorified Markov chain generators that didn't actually understand code and couldn't produce anything novel. I started this project on a lark, fully expecting the AI to produce terrible code for me to laugh at. And then, uh... the code actually looked pretty good. Not perfect, but I just told the AI to fix things, and it did. I was shocked.
To emphasize, this is not "vibe coded". Every line was thoroughly reviewed and cross-referenced with relevant RFCs, by security experts with previous experience with those RFCs. I was trying to validate my skepticism. I ended up proving myself wrong.
Again, please check out the commit history -- especially early commits -- to understand how this went.
See the many comments on HN about this
It took an expert and a whole team of software engineer to code "with IA" something well defined, well known, and with already lots of implementation in other languages
Glorious indeed, we have created a new kind of printer ! (irony inside)
(in conclusion, the above post from cloudflare is an ad)
> TO EMPHASIZE THIS WAS NOT VIBE CODED
an expert using high powered autocomplete to successfully take a shortcut towards a well defined problem is not evidence of "AGI" or "the singularity"
belief in "the singularity" is a religious belief. you people sound like evangelicals talking about the goddamned rapture
Sorry, yes, talk of “AGI” or “the singularity” is silly and I should’ve distanced myself from that bit of the post. But the use of AI to write durable code is happening now, like it or not.
It can only be categorized as durable after careful revision from human experts and extensive testing. This is what the article is about.
[dead]
See also centuries of chrysopoeia: https://en.wikipedia.org/wiki/Chrysopoeia.
To be fair, they did synthesize gold in 1984, "although the production cost is estimated to be a trillion times the market price of gold."
[dead]
The question remains, 'can it be done'.
I write deplorable code , personally