Trust the AI, says new coding manifesto by Kim and Yegge "Accept All. Always. Don't read the diffs anymore." This is OpenAI co-founder Andrej Karpathy's vision of "vibe coding" as he described on X in February – a new approach where developers surrender control to AI agents and simply trust the vibes. Vibe Coding by Gene Kim and Steve Yegge Vibe Coding by Gene Kim and Steve Yegge …
My problem with this is exactly that these individuals are incredibly experienced. They can happily vibe code to their hearts content because they can spot when the AI gets it wrong.
What they are advocating though fails to give the "new generation" the skills and experience to know this for the future - "just trust the AI and all will be fine". To build something, you need successful foundations - you need to put in the hard work to actually understand something, or you'll eventually find out the hard way.
Ironically, on the same day, we also seem to be discussing a colossal failure at Amazon, potentially because people didn't have the skills and experience to see why something was failing and to correct it.
>> Not at all. Vibe coding is more akin to asking someone who knows nothing about writing software to get pieces of code from all over the internet, put them together, and hope it does what you want it do
Unfortunately, even before LLMs, bosses tended to push this sort of crap approach. Every time I read garbage like "give coding", I'm more relieved that I'm retired and don't have to push back against such unprofessional stupidity.
Who knows, maybe in my 80s or 90s it will all have fallen apart so badly that it might be tempted to show the retards (see Marching Morons by Kornbluth) how to do it right. On second thought, nah.
This post has been deleted by its author
Those were my thoughts as well. We have a proper nerdy greybeard in our department. He did FOSS development on the side, knows the people, is rooted in IT, has been working on databases since around the mesothitic period (i.e. he is even older than I am). He is incredibly experienced and likes AI to just spin up a quick project for him - because he knows how things should behave and what to watch out for. Experience is needed. A lot of it.
I would compare it to car maintenance: if you understand how an engine works and have wrenched on everything starting from, say, the 80s when things were pretty simple and mechanical or electromechanical and you could see and understand the parts and then look at the output of one of the new error-code-showing computer-on-wheels-management-thingys (like you now need in every shop) you can tell if the error code makes sense or not. You will not just continue swapping parts in the hope the error goes away. But you need that experience and deep understanding.
That's why I could fix my old Golf (simple, mechanical) but not the new Astra (computer on wheels where I would need to understand how stuff actually works because wtf is going wrong is hidden and abstracted away), that's why I can write half-decent code but should not spin up a ton of vibe-coded containers (or feel nervous to do so).
I think it likely both of them are are in the business of building and debugging AI software tools, and therefore - as they should be - eating the dogfood as they bake it.
Which is fine! But it also means that the book is not an unbiased source, and that their enthusiastic usage is not the proof of a finished safe and productive tool they would have you believe.
I read the "Phoenix Project" by Kim et al and thought it was an interesting take on explaining DevOps.
This seems like a bandwagon jumping piece of nonsense that they've thrown together quickly to get a few extra $$$. If vibe coding works, take the credit. If it doesn't blame the AI. What a depressing surrender of IT to the lowest common slop denominator.
Icon because Kim and Yegge clearly don't believe that there is such a thing as IT anymore - as long as the vibe's right who cares whether it works or not as long as you personally shovel up the dollars.
South Park does seem to be struggling to get episodes out at the moment doesn't it ? Missing schedules, and just now cutting S27 short, turning episode s27e6 into s28e1. The official line seems to be that "its all too difficult keeping up with the madness so we need a bit longer" and "we always intended just 5 episodes per season, just forgot to tell anyone". Take that as you will. I'm trying to avoid drawing conclusions of conspiracy, but it does invite speculation about hard meetings at Paramount.
It would depend what this $1.5bn contract between Paramount and South Park was... 5 more years or 5 more seasons. It was generally reported as 5 more years but maybe people assumed there would be one season per year. It looks like Paramount is trying to cut its exposure to El Presidente's wrath.
The swiftian satire is the current administration in the US of A.
The book is just a variation on the current messaging that 'AI' works you are just 'doing it wrong !!!'.
Eventually we will have a real large scale disaster caused by 'AI' and some idiot believing the message that 'AI' works.
"You can only expand your boundaries by pushing the limits of what you can do" ... nice concept to align with ... BUT in real life that means someone pays usually by losing a limb or two, if lucky, else lives are lost !!!
Not a problem though as it will not be anyone 'important', as far as the 'AI' behemoths are concerned.
:)
Foreword by Dario Amodei, CEO of Anthropic who said 6 months ago that in 6 months 80% of code would be written by AI's, who's one shining hope for his company making more money than they currently spend on AWS and Google Cloud is to get *everyone* hooked on AI coding so he can then raise prices by 2, 5 or 10x to stop losing about $1 every time someone actually prompts Claude Code to do anything.
No, no hidden agenda here.
Actually does the book get into how much turning into a Vibe coder costs in terms of the required subscription to something like Cursor or Claude Code to access enough tokens to get these tools to do anything useful?
I saw one guesstimated number doing the rounds that some company's are looking at AI coding bills that are in the ballpark of a senior developer's annual salary per developer using AI tools. In fact there's this story in El Reg https://www.theregister.com/2025/08/15/are_you_willing_to_pay/
Also plug for Ed Zitron's latest wheresyouredat newsletter/screed/magnum opus where he dives in and appears to show that Anthropic are currently spending 100+ % of their gross revenue on their AWS bill and maybe another 25% of revenue on GCP never mind other incidental bills like salarys, office rental, sales, marketing, etc and that all thats keeping them going is an endless series of funding rounds which is always a sustainable way to run a business.
This post has been deleted by its author
"Part three turns to the confusing profusion of vibe coding tools and agents, and how the classic developer loop – code, compile, run, test, debug – changes in vibe coding to one that starts with defining a subtask and continues with an AI conversation, AI-generated plan, and so forth. The importance of tests is repeatedly emphasized."
Here's ChatGPT's authentic vibe style rewrite of the above:
"Part -5+8J turns to whatever camel-case lunatic product name they've come up with - probably something deeply "meaningful" written in ancient Sanskrit. Who cares, we'll be telling you it's completely passe in a fortnight anyway (by which point all of the developers in the world will have turned into pixies - or was it cats? WHATEVER!). With vibe coding, you just need to ask some half-arsed question about something or other and the magic wiffledust will turn it into a fully functioning app that will make you millions of dollars in a month. Honest. Testing is for losers anyway - who cares if it works as the code it produces is so impenetrable nobody can understand what, how or whether it does anything- JUST GIVE ME THE MONEY NOW YOU SAP!"
So what they're saying is if you know exactly what to say to get the output you want, have enough relevant experience and training to polish up the results or just do it yourself from scratch when needed, and can properly identify when it's intentionally behaving badly and producing highly undesirable results, vibe coding is perfectly fine.
Yep, sounds great. Coding for the masses is finally here!
</sarcasm>
I used to really enjoy Steve Yegge's blog rants, he's a very experienced engineer and a good storyteller. Him taking AI development seriously was one of the things that persuaded me there might be a baby in the AI bathwater.
Now that I have worked with it more, the baby looks ever smaller. If anything it appears to be taking on the characteristics of a damp rat.
All of which to say I'm a little disappointed in seeing someone whose work I used to enjoy move into selling a technology where I can't perceive the benefits that he does, and where I'm consequently no longer sure he deserves the benefit of the doubt.
