The one thing that gets me is - Boris must know what everyone is thinking when he says he merges 300 MR’s per day, but I think everyone knows it doesn’t mean what he’s implying - it just can’t. He can’t read 300 open source contributions per day to determine if they fit into the spirit of Claude Code. Even if he’d fully automated the testing and integration process. And 300 people per day submitting contributions?
He could mean a few other thingsc one would be, he has like 20 version tags and he merges 15 features into each version (does he explicitly say merge to main?)
The other thing he could mean is he has like a software engineering agent and that software engineering agent like loops through GitHub issues and his personal notepad and maybe uses a few different branches to test things out and build I dunno adversarially, running all sorts of experiments.
Which would be genuinely cool! But using Mr’s to say tweak bunch of variables back and forth isn’t what 300 MR’s implies.
But then the ultimate question is - it may be cool to fully automate a software engineering agent, and certainly is the type of research Anthropic and someone of Boris’ stature (and pay level) should be working on. But is it efficient?
I think this guy is using AI differently than me. Since Opus 4.6 and GPT 5.3, I have been able to absolutely crush my coding work. Boris might be embellishing how he ONLY writes loops, but for the most part I am just handing off planning docs to Claude or GPT and they implement it with like 95% accuracy.
A lot of you don't want to hear it but this is a user issue.
I read this a lot and it is just very foreign to me. I use AI systems in software work all day seven days a week and my job has become simultaneously more interesting and more difficult because I scale the ambition up until it's hard again.
Isn't anything else a surrender to irrelevance? I agree that many coding tasks that were previously effort intensive are now not effort intensive, but there's no ceiling I'm aware of on how correct and performant and economical and capable software can be short of saturating the hardware.
And the emergence of agentic intelligence at scale demands new regimes of performance and correctness and economy like maybe nothing else ever has.
I have an anecdote related to TUI flickering in that my TUI library had a flickering problem because it was doing more than 10k FPS, and so I had to lock the buffer swap to the vsync to stop it tearing.
AI coding didn't make more React too cheap to meter, it made notcurses bound into Trinity-inspired deterministic replay event substrate over io_uring possible.
I am in no way surprised a sufficient waterfall method passed to Claude code could result in a completely accurate application. But most applications aren’t built via waterfall for all the reasons.
Also agents are just loops. So if you use Claude Code you are doing. Everything with a loop. So I do believe him but Im not entirely grokking the flex.
> A lot of you don't want to hear it but this is a user issue.
Hmm people say this all the time in these discussions but idk if I buy that it's really a "user issue". It's really, really not hard to "use" agentic ai. It literally involves instructing an llm to do things in natural language. Anyone who knows how to code and speak a language can do this. As you yourself seem to believe, even people who don't know how to code can do this. I just don't think it's possible that THAT many people are having an issue typing some words to instruct an llm to write some code. Maybe the issue is more the type of software you are working on vs the type of software that other people are working on. I don't know, I just don't think "skill issue" is really a valid argument here...
Edit: for the record, I think what Opus can produce is extremely impressive. But I still am not really close to letting write 100% of code I write. And I think that is true for a lot of people, not just me. It still generates (sometimes obvious) bugs. Until that stops, the statement "coding is solved" is objectively false, which (I think?) is largely the point of the video.
If Anthropic has some of the best and most highly paid software engineers on the planet working on a simple program (terminal app) with virtually unlimited tokens for the strongest coding model, and they still ship a sloppy buggy mess, what does that say about the quality of code you are outputting?
A lot of you don't want to hear it, but you aren't doing better than Anthropic so unless your use case is ridiculously simple, Claude Code is the ceiling for what you're creating, and if what you're doing is at all complex, the ceiling is much much lower.
> just handing off planning docs to Claude or GPT and they implement it with like 95% accuracy.
Do you have any publicly available demonstrations of this claim?
> A lot of you don't want to hear it
That there are skill differences in the use of technology? On the contrary this knowledge makes me suspicious of undocumented claims like yours.
> this is a user issue.
Another claim I wish was quantified. With all the billions invested I assumed this would naturally come to exist. I may have just missed it. Any pointers?
> Do you have any publicly available demonstrations of this claim?
Yeah I mean for example I wrote up a new audio mixer application for TDE using basically claude and just saying - hey rewrite the old ALSA one with Pulse/Pipewire.... its awesome. I dont know how it works.,
Sorry, but context rot is real, and I’d be curious how your code is playing out in the real world. Is it shipping? Is it a known product with stable docs? Is it greenfield?
Aspects of coding are faster certainly, but oh gosh can it get very wrong very fast when things go sideways, and with everyone using it, the chaos factor compounds into a near halt.
I am definitely hyped on it and have 10-20 Codex and Claude Code terminals open, but I do wonder what you're building and with who that you would say 95% accuracy. I get so frustrated with Codex inventing new ways to do something every time it compacts or start a new session.
"You're right, I shouldn't have done that" Ya think?!
Yup. The only caveat I'd add is that I'm using an alternate account to agree with people who say that AI coding has been amazing, because there is a seemingly a good chunk of people who dislike it and it will be met with downvotes. Also because my real account has my real name in the profile along with projects I work on, and a simple search could reveal my pro-AI coding views and these same folks who downvote could also be a future interviewer.
I think the world changed. And it's changed for the good. AI is a tool, and we should not be afraid of this tool for the coding world. I am only speaking about coding, I'm not speaking about other uses of AI, just so that we're clear on the scope of what I mean by good change.
For the first time, I see people who had all these ideas finally bring them to reality and watch it blossom. They wanted to build something to share with their communities, but the walls were too high. Too much gatekeeping. Too much of thinking that programming was a task for the elite few and not for the masses. Along the way, we all forgot that we build tools for people. And having an additional tool help us make better tools for people is a win. Just below this comment, I see people talking about dementia, "lots more generated code, almost all of it garbage", "future where garbage software".
I think the only delusional ones are the idea that humans were better at coding. Have you never had to work on an older project? One that you did not have to start fresh on? Or did you come into either one and go "wow, this is perfect! everything is so beautiful!" Do you seriously consider your fresh project (that didn't use AI) to be the best most perfect beautiful code ever?
The fact is that nobody cares. People want to use good things and have fun with their lives. They're not worried about whether you wrote a method that parses some strings beautifully or did it with a one-liner. That never mattered, and I think a lot of you can't let go of that world view change and instead lash out at people who simply embrace that programming was simply a tool, not some elite special skill. And we're going back to those beliefs. It's done. It's over. Get over it.
Agreed. I think the opinion on AI is split into two camps, people who's enjoyment from programming came from writing the code and people who like building things. It's really undeniable at this point that AI has changed the job and I really enjoy it now more than ever, I can come up with an idea, guide an LLM through the steps to build it and have it be a real thing faster than I ever could have imagined.
Yeah LLMs aren't perfect, there is back and forth along the way and if you just let it loose you are going to end up with slop but I feel like we can achieve better quality now in a shorter amount of time using the tool properly. I'm not sure if I am just naive but I am really excited about the possibilities now and have been spending more time than ever building what I want. I used to think that writing the code was the enjoyable part for me but I think it was just building things.
I empathize with people in the other camp who got into it for the love of the code and now that part of the job is being taken away but I think it would benefit them to be honest about LLMs and try and work out a path forward here rather than just "my function is better than an LLM one, LLMs are just slop machines"
Well said. Most of my career I made a trade-off and that trade-off was that I would much rather spend my free time outside in nature than on implementing my wild ideas which I always recognized would take considerable time. I'd maybe take 1% of my ideas anywhere. Now I can play with ideas while I'm out on the trail and turn those into something I can test within a couple hours, it's the most fun I've had on computers since the very beginning.
Their apparent inability to get the basics right makes me severely doubt their claims of self-improving AI. The humans at Anthropic wouldn't know improvement if it landed on their lap and started twerking, and AI cannot do a job without strong human intervention into what the goals and guardrails actually are.
I'm kind of reminded of when Microsoft claimed it took a team of Ph.D.s to write a terminal application that updated at 60fps, and then Casey Muratori did it over a weekend. And this was before AI was writing code in earnest; when LLM-induced brainrot really sets in, civilization is in for a world of fresh hurt: lots more generated code, almost all of it garbage. And the promised AI crossover point where it becomes AGI, or indistinguishable from for software design purposes, recedes into the infinite future.
Reading this it occurs to me that this timeline may be moving toward a future where garbage software, garbage information etc. have become the norm for so long that the number of people who can distinguish trash from quality, or signal from noise, has become negligible.
A true era of ignorance, looking like an ocean of nonsense in which no one can really navigate as it is ungrounded in reality.
Idiocracy presents a naively gentle positive version of such future but there are many darker ones possible.
I watched the video and I wish I could get those 13 minutes of my life back.
He could have done it in 13 seconds instead of 13 minutes: "Anthropic is lying about the effectiveness of agentic loops because there's this one screen flicker bug in Claude Code that took a year to fix."
Yeah, like when United Airlines claims a plane can fly 300 people 6,000 miles they are lying to you.
I can prove they're lying to you because people have been complaining about uncomfortable seats and flight delays for literally decades and those issues still aren't fixed.
I write little bits of code then test them. If the screen flickers i wouldnt continue until it is solved. If ive missed it and have to hear it from (a) user(s) i would kinda like to fix it imediately. Not always possible but truly annoying things have priority over new things.
If i had unlimited developers at my command. All many times as fast as me.... how can i keep the problem? It would take some huge effort to keep.
The unlimited number of devs would talk about it an unlimited number of times. Not fixing it would be very expensive that way.
He spent that much time and you still misunderstood the direct message and missed the subtext.
The lie is coding is solved, the proof is they had an outstanding coding issue they were working on for over a year while saying coding is solved. There’s a great number of other issues with their own software that disprove their premise, but you only need one counter example to disprove something.
And because you missed it, the subtext was they want you to use loops not because they work but because they burn lots of tokens thus making them more money.
> I'm kind of reminded of when Microsoft claimed it took a team of Ph.D.s to write a terminal application that updated at 60fps, and then Casey Muratori did it over a weekend.
This is the same Microsoft that is now rewriting the TypeScript type checker, parser and its developer tools in Go after realizing that the bottleneck was...the performance of TypeScript itself, which is a basic compiled vs interpreted difference.
> And this was before AI was writing code in earnest; when LLM-induced brainrot really sets in, civilization is in for a world of fresh hurt: lots more generated code, almost all of it garbage.
Some folks using LLMs wouldn't realize why it makes zero sense to use TS / JS for building performant and optimal applications. This is why people were experiencing significant rendering bugs in terminal apps (they are not designed for that) and slow starts with Claude Code, which was completely vibe coded with Ink.
If you don't understand the basic fundamentals of what you are working on with LLMs and bugs are creeping up left and right, then you are just sinking in your own comprehension debt.
To be clear: I'm an extremely experienced software developer. I use AI daily. I'm currently working on training a DNN that's used in production by enterprise companies around the globe.
But there's absolutely no doubt that AI is being deployed as a weapon by the capitalist class against everyone else, and that's only going to amplify and get much, much worse. The issue is not necessarily the technology itself. The issue is how society permits it to be used.
The one thing that gets me is - Boris must know what everyone is thinking when he says he merges 300 MR’s per day, but I think everyone knows it doesn’t mean what he’s implying - it just can’t. He can’t read 300 open source contributions per day to determine if they fit into the spirit of Claude Code. Even if he’d fully automated the testing and integration process. And 300 people per day submitting contributions?
He could mean a few other thingsc one would be, he has like 20 version tags and he merges 15 features into each version (does he explicitly say merge to main?)
The other thing he could mean is he has like a software engineering agent and that software engineering agent like loops through GitHub issues and his personal notepad and maybe uses a few different branches to test things out and build I dunno adversarially, running all sorts of experiments.
Which would be genuinely cool! But using Mr’s to say tweak bunch of variables back and forth isn’t what 300 MR’s implies.
But then the ultimate question is - it may be cool to fully automate a software engineering agent, and certainly is the type of research Anthropic and someone of Boris’ stature (and pay level) should be working on. But is it efficient?
I guess yes hems talked about this:
https://karozieminski.substack.com/p/boris-cherny-claude-cod...
I think this guy is using AI differently than me. Since Opus 4.6 and GPT 5.3, I have been able to absolutely crush my coding work. Boris might be embellishing how he ONLY writes loops, but for the most part I am just handing off planning docs to Claude or GPT and they implement it with like 95% accuracy.
A lot of you don't want to hear it but this is a user issue.
I read this a lot and it is just very foreign to me. I use AI systems in software work all day seven days a week and my job has become simultaneously more interesting and more difficult because I scale the ambition up until it's hard again.
Isn't anything else a surrender to irrelevance? I agree that many coding tasks that were previously effort intensive are now not effort intensive, but there's no ceiling I'm aware of on how correct and performant and economical and capable software can be short of saturating the hardware.
And the emergence of agentic intelligence at scale demands new regimes of performance and correctness and economy like maybe nothing else ever has.
I have an anecdote related to TUI flickering in that my TUI library had a flickering problem because it was doing more than 10k FPS, and so I had to lock the buffer swap to the vsync to stop it tearing.
AI coding didn't make more React too cheap to meter, it made notcurses bound into Trinity-inspired deterministic replay event substrate over io_uring possible.
https://youtu.be/YqgEtpJ8tGI?feature=shared
He's discussing Anthropic struggling to fix an issue with their own product. He's not the one struggling.
You have planning docs?
I am in no way surprised a sufficient waterfall method passed to Claude code could result in a completely accurate application. But most applications aren’t built via waterfall for all the reasons.
Also agents are just loops. So if you use Claude Code you are doing. Everything with a loop. So I do believe him but Im not entirely grokking the flex.
> A lot of you don't want to hear it but this is a user issue.
Hmm people say this all the time in these discussions but idk if I buy that it's really a "user issue". It's really, really not hard to "use" agentic ai. It literally involves instructing an llm to do things in natural language. Anyone who knows how to code and speak a language can do this. As you yourself seem to believe, even people who don't know how to code can do this. I just don't think it's possible that THAT many people are having an issue typing some words to instruct an llm to write some code. Maybe the issue is more the type of software you are working on vs the type of software that other people are working on. I don't know, I just don't think "skill issue" is really a valid argument here...
Edit: for the record, I think what Opus can produce is extremely impressive. But I still am not really close to letting write 100% of code I write. And I think that is true for a lot of people, not just me. It still generates (sometimes obvious) bugs. Until that stops, the statement "coding is solved" is objectively false, which (I think?) is largely the point of the video.
This guy has been against AI from the very onset, and no matter what happens with AI going forward I think he'll always poopoo it.
If Anthropic has some of the best and most highly paid software engineers on the planet working on a simple program (terminal app) with virtually unlimited tokens for the strongest coding model, and they still ship a sloppy buggy mess, what does that say about the quality of code you are outputting?
A lot of you don't want to hear it, but you aren't doing better than Anthropic so unless your use case is ridiculously simple, Claude Code is the ceiling for what you're creating, and if what you're doing is at all complex, the ceiling is much much lower.
> just handing off planning docs to Claude or GPT and they implement it with like 95% accuracy.
Do you have any publicly available demonstrations of this claim?
> A lot of you don't want to hear it
That there are skill differences in the use of technology? On the contrary this knowledge makes me suspicious of undocumented claims like yours.
> this is a user issue.
Another claim I wish was quantified. With all the billions invested I assumed this would naturally come to exist. I may have just missed it. Any pointers?
> Do you have any publicly available demonstrations of this claim?
Yeah I mean for example I wrote up a new audio mixer application for TDE using basically claude and just saying - hey rewrite the old ALSA one with Pulse/Pipewire.... its awesome. I dont know how it works.,
Sorry, but context rot is real, and I’d be curious how your code is playing out in the real world. Is it shipping? Is it a known product with stable docs? Is it greenfield?
Aspects of coding are faster certainly, but oh gosh can it get very wrong very fast when things go sideways, and with everyone using it, the chaos factor compounds into a near halt.
When you hand something off to Claude code, the harness is doing lots of different sessions it’s not a one shot.
I am definitely hyped on it and have 10-20 Codex and Claude Code terminals open, but I do wonder what you're building and with who that you would say 95% accuracy. I get so frustrated with Codex inventing new ways to do something every time it compacts or start a new session.
"You're right, I shouldn't have done that" Ya think?!
Yup. The only caveat I'd add is that I'm using an alternate account to agree with people who say that AI coding has been amazing, because there is a seemingly a good chunk of people who dislike it and it will be met with downvotes. Also because my real account has my real name in the profile along with projects I work on, and a simple search could reveal my pro-AI coding views and these same folks who downvote could also be a future interviewer.
I think the world changed. And it's changed for the good. AI is a tool, and we should not be afraid of this tool for the coding world. I am only speaking about coding, I'm not speaking about other uses of AI, just so that we're clear on the scope of what I mean by good change.
For the first time, I see people who had all these ideas finally bring them to reality and watch it blossom. They wanted to build something to share with their communities, but the walls were too high. Too much gatekeeping. Too much of thinking that programming was a task for the elite few and not for the masses. Along the way, we all forgot that we build tools for people. And having an additional tool help us make better tools for people is a win. Just below this comment, I see people talking about dementia, "lots more generated code, almost all of it garbage", "future where garbage software".
I think the only delusional ones are the idea that humans were better at coding. Have you never had to work on an older project? One that you did not have to start fresh on? Or did you come into either one and go "wow, this is perfect! everything is so beautiful!" Do you seriously consider your fresh project (that didn't use AI) to be the best most perfect beautiful code ever?
The fact is that nobody cares. People want to use good things and have fun with their lives. They're not worried about whether you wrote a method that parses some strings beautifully or did it with a one-liner. That never mattered, and I think a lot of you can't let go of that world view change and instead lash out at people who simply embrace that programming was simply a tool, not some elite special skill. And we're going back to those beliefs. It's done. It's over. Get over it.
Agreed. I think the opinion on AI is split into two camps, people who's enjoyment from programming came from writing the code and people who like building things. It's really undeniable at this point that AI has changed the job and I really enjoy it now more than ever, I can come up with an idea, guide an LLM through the steps to build it and have it be a real thing faster than I ever could have imagined.
Yeah LLMs aren't perfect, there is back and forth along the way and if you just let it loose you are going to end up with slop but I feel like we can achieve better quality now in a shorter amount of time using the tool properly. I'm not sure if I am just naive but I am really excited about the possibilities now and have been spending more time than ever building what I want. I used to think that writing the code was the enjoyable part for me but I think it was just building things.
I empathize with people in the other camp who got into it for the love of the code and now that part of the job is being taken away but I think it would benefit them to be honest about LLMs and try and work out a path forward here rather than just "my function is better than an LLM one, LLMs are just slop machines"
Well said. Most of my career I made a trade-off and that trade-off was that I would much rather spend my free time outside in nature than on implementing my wild ideas which I always recognized would take considerable time. I'd maybe take 1% of my ideas anywhere. Now I can play with ideas while I'm out on the trail and turn those into something I can test within a couple hours, it's the most fun I've had on computers since the very beginning.
Very well said yourself! I frequently tell my friends (both in tech and polar opposite non-tech) that 3 things changed the tech world for me:
1. Internet 2. Smartphone 3. AI coding
All 3 were "WOW" moments for me.
Can you give me your use case here? I have not gotten around to trying loops in Claude code but have started to notice the hype.
Pretty much the same point I made: https://news.ycombinator.com/item?id=48403908
Their apparent inability to get the basics right makes me severely doubt their claims of self-improving AI. The humans at Anthropic wouldn't know improvement if it landed on their lap and started twerking, and AI cannot do a job without strong human intervention into what the goals and guardrails actually are.
I'm kind of reminded of when Microsoft claimed it took a team of Ph.D.s to write a terminal application that updated at 60fps, and then Casey Muratori did it over a weekend. And this was before AI was writing code in earnest; when LLM-induced brainrot really sets in, civilization is in for a world of fresh hurt: lots more generated code, almost all of it garbage. And the promised AI crossover point where it becomes AGI, or indistinguishable from for software design purposes, recedes into the infinite future.
Reading this it occurs to me that this timeline may be moving toward a future where garbage software, garbage information etc. have become the norm for so long that the number of people who can distinguish trash from quality, or signal from noise, has become negligible.
A true era of ignorance, looking like an ocean of nonsense in which no one can really navigate as it is ungrounded in reality.
Idiocracy presents a naively gentle positive version of such future but there are many darker ones possible.
Kali Yuga, indeed.
https://en.wikipedia.org/wiki/Kali_Yuga
Customers abandon companies that fail.
Remember how they used Brando to water plants and it kills them? Eventually mistakes break critical systems and you fail.
Ignorance is too generous a word. This is epistemic collapse.
The pralaya is not far away. Soon Shiva will begin his dance. I give us until 2036.
Pralaya happens after complete age of Brahma. That would be 100 Brahma years. Or 311.04 trillion human years. We are in the 51st year of Brahma.
Kali Yuga lifespan is 432,000 years. Of which we are 4000+ years into it. So that's another 428,000 years of hell on Earth.
Yeah like he said, not far away :D
Perhaps, but people don't care that much about a bug with Instagram. Once critical infrastructure starts failing people will take notice.
Agreed. Watch for a rise in cases of early onset dementia over the next few decades.
No, we'll just find harder problems.... Coding is boring now
"The Feeling of Power" by Arthur C. Clarke
Isaac Asimov, but yes.
argh
Pretty much the same point I made https://news.ycombinator.com/item?id=48500537
I watched the video and I wish I could get those 13 minutes of my life back.
He could have done it in 13 seconds instead of 13 minutes: "Anthropic is lying about the effectiveness of agentic loops because there's this one screen flicker bug in Claude Code that took a year to fix."
Yeah, like when United Airlines claims a plane can fly 300 people 6,000 miles they are lying to you.
I can prove they're lying to you because people have been complaining about uncomfortable seats and flight delays for literally decades and those issues still aren't fixed.
I write little bits of code then test them. If the screen flickers i wouldnt continue until it is solved. If ive missed it and have to hear it from (a) user(s) i would kinda like to fix it imediately. Not always possible but truly annoying things have priority over new things.
If i had unlimited developers at my command. All many times as fast as me.... how can i keep the problem? It would take some huge effort to keep.
The unlimited number of devs would talk about it an unlimited number of times. Not fixing it would be very expensive that way.
[dead]
He spent that much time and you still misunderstood the direct message and missed the subtext.
The lie is coding is solved, the proof is they had an outstanding coding issue they were working on for over a year while saying coding is solved. There’s a great number of other issues with their own software that disprove their premise, but you only need one counter example to disprove something.
And because you missed it, the subtext was they want you to use loops not because they work but because they burn lots of tokens thus making them more money.
One unresolved bug does not disprove anything.
The point is if "coding is solved" was true, there would not be any unresolved bugs.
And saying coding is solved doesn't prove anything either.
> I'm kind of reminded of when Microsoft claimed it took a team of Ph.D.s to write a terminal application that updated at 60fps, and then Casey Muratori did it over a weekend.
This is the same Microsoft that is now rewriting the TypeScript type checker, parser and its developer tools in Go after realizing that the bottleneck was...the performance of TypeScript itself, which is a basic compiled vs interpreted difference.
> And this was before AI was writing code in earnest; when LLM-induced brainrot really sets in, civilization is in for a world of fresh hurt: lots more generated code, almost all of it garbage.
Some folks using LLMs wouldn't realize why it makes zero sense to use TS / JS for building performant and optimal applications. This is why people were experiencing significant rendering bugs in terminal apps (they are not designed for that) and slow starts with Claude Code, which was completely vibe coded with Ink.
If you don't understand the basic fundamentals of what you are working on with LLMs and bugs are creeping up left and right, then you are just sinking in your own comprehension debt.
I mean you can go through Boris' history here on hn to see he is a liar.
It goes beyond lying. It's kind of war, and they're the aggressor.
Everyone else needs to start treating them that way, or you're going to regret it once you realize what's actually happening.
Please please tell us so we're prepared. sad puppy eyes
There's plenty of discussion of this if you look for it. Some examples:
"Sanders Releases Report on Big Tech Oligarchs’ War Against Workers, Warns AI Could Eliminate Nearly 100 Million U.S. Jobs": https://www.help.senate.gov/dem/newsroom/press/news-sanders-...
"How the American Oligarchy Went Hyperscale": https://www.motherjones.com/politics/2026/04/american-oligar...
"AI Is a Threat to Everything the American People Hold Dear": https://www.sanders.senate.gov/op-eds/ai-is-a-threat-to-ever...
To be clear: I'm an extremely experienced software developer. I use AI daily. I'm currently working on training a DNN that's used in production by enterprise companies around the globe.
But there's absolutely no doubt that AI is being deployed as a weapon by the capitalist class against everyone else, and that's only going to amplify and get much, much worse. The issue is not necessarily the technology itself. The issue is how society permits it to be used.
What does this have to do with the video?
[dead]
[dead]
[flagged]