So long as perceived LLM skill is still "spiky" - e.g. within a domain, still showing relatively high variation in ability (often depending on the task or user, to be fair), people will continue to dismiss it
I find these water-against-a-rock literary tones so tedious. Even the writer always seems to have to go back and put some of it in BOLD TEXT, supposedly highlighting the main ideas, but really optical affordances.
The truth of this seems much more banal. Computing has become a major drag. There have been tens of thousands of libraries that reinvent the wheel. Every operating system has become a toy. All major language systems have an absurd learning curve. Each important application is fortified by a giant corporation. Social media is self-important pop babble.
LLMs are surprisingly good at dealing with complex systems. I can fire one up and ask, for example, why this Swift code is not compiling. But why doesn’t my Swift editor explain that problem? Why is it a confusing question at all? The entire system was built from the ground up at enormous expense. Why do I seek outside help?
Our computing is full of whizzy animations and pointless Victorian ironmongery. All meaningless. AI is medicine, not the cure.
Because if we engineered these to work, then we'd be out of a job because the problems would be solved and we could not sell the same service or software 1000 times.
Management is going to quickly start bisecting human engineers along lines of maximalists and minimalists. The minimalists will all be let go. A few bad things will happen. A few systems will strain under the pressure but itll be “worth it” in the same way that its cheaper to pay lawsuits than do a recall of a plane.
We arent innovating in other areas that might soften the blow. We dont have good support systems, social security, healthcare, or even demands in other areas. How many engineers are going to be plumbers and construction workers?
Jokes aside it should be noted that the author is a founder and ceo of an AI company, not to mention an active investor in the sector. (All disclosed in his "about" page)
It should be noted that the author doesn't shy away from that, and that his argument is convincing. He notes that while he is in the AI field, the actual cutting edge of AI development is done by a far smaller group of companies and researchers than the broader AI industry which includes his company.
Was there something specific in the article you found unconvincing, or that directly counters an experience you've had with AI?
What’s the point in using these tools if they’re gonna replace us in a few years? It’s weird the author says that but then his conclusion is basically “go spend money on stuff I’m invested in”.
Covid comparison is apt. I remember being insanely scared in Jan 2020 when those videos of Chinese people dropping dead were coming out (and being shamed by most of my peers etc). Few months later it was starting to become obvious it was really only a major risk if you were old or infirm, but the rest of the world had took awhile to catch up.
AI’s big and gonna change stuff - and like COVID probably for the worse - but we’re in a poorly understood hype cycle right now.
Yep! They still remain stupid (in the intelligence sense, not the pejorative sense) tools which have no practical value to anyone with decent skills. But the people who have a financial interest in hype are still trying to convince everyone "it's totally different now bro, use the latest model bro". It's so tiresome.
I am not having the exact same experience as the author--Opus 4.6 and Codex 5.3 seem more incremental to me than what he is describing--but if we're on an exponential curve, the difference is a rounding error.
4 months ago, I tried to build an application mostly vibe-coded. I got impressively far for what I thought was possible, but it bogged down. This past weekend, my friend had OpenClaw build an application of similar complexity in a weekend. The difference is vast.
At work, I wouldn't say I'm one-shotting tasks, but the first shot is doing what used to be a week's work in about an hour, and then the next few hours are polish. Most of the delay in the polish phase is due to the speed of the tooling (e.g. feature branch environment spin up and CI) and the human review at the end of the process.
The side effects people report of lower quality code hitting review are real, but I think that is a matter of training, process and work harness. I see no reason that won't significantly improve.
As I said in another thread a couple days ago, AI is the first technology where everyone is literally having a different experience. Even within my company, there are divergent experiences. But I think we're in world where very soon, companies will be demanding their engineering departments converge to the lived experience of the people who are seeing something like the author. And if they can find people who can actuate that reality, the folks who can't are going to see their options contract precipitously.
This is a solid assessment of whats here and what is in front of us. Broad brush stroke dismissals aside, we are here. Evolve or Perish. AI is like unchecked fire, but make no mistake fire is very powerful once it was harnessed. AI leans more supplemental vs incremental than prior major tech shifts and that's worth noting. It will be the same for other sectors and verticals over time. The markets view software eats the world is being eaten by new software.
> You can describe an app to AI and have a working version in an hour. I'm not exaggerating. I do this regularly. If you've always wanted to write a book but couldn't find the time or struggled with the writing, you can work with AI to get it done.
Interesting. So you regularly make new apps in 1 hour each.
How is that the same as...writing a book? Did you mean write several short stories? Or are we talking non-fiction?
Every once in a while, I try LLMs just to see how improvement is going.
Yesterday I had to explain to Opus what the color white is and what "bottom right" means after it declared problems fixed, repeatedly, that a literal preschooler would have been able to tell were absolutely unchanged from the original problem description.
I am still waiting for this world of redundant programmers I've been hearing about for years.
I use FreeBSD. When talking to LLMs, they insist on giving me code in bash. Bash is not native to FreeBSD (though you can get it and use it). I correct them, and of course they apologize, but another day continue to use bash in other questions.
Considering the rate of improvement of these LLMs, wait for a month or two and then you may not even need an os, let alone some obsecure piece of software (shell).
Except the models don’t actually compute anything other than text generation. The entire way they interact with computers is through the shell or other api layers on the OS
PhD physicist (Stanford/SLAC), Research Software Engineer doing low-level systems work in C/C++ and LLM research. Not a founder or investor — just a practitioner.
One data point for this thread: the jump from Opus 4.5 to 4.6 is not linear. The minor version number is misleading. In my daily work the capability difference is the largest single-model jump I've experienced, and I don't say that casually — I spent my career making precision measurements.
I keep telling myself I should systematically evaluate GPT-5.3 Codex and the other frontier models. But Opus is so productive now that I can't justify the time. That velocity of entrenchment is itself a signal, and I think it quietly supports the author's thesis.
I'm not a doomer — I'm an optimist about what prepared individuals and communities can do with this. But I shared this article with family and walked them through it in detail before I ever saw it on HN. That should tell you something about where I think we are.
If you use Claude Code, it will take you half a day to learn to use Codex, and like 30 minutes to start being productive in it. The switching cost is almost zero. Just go test out GPT 5.3, there is no reason not to
It's a bit more than zero, because I have substantial tooling around Claude Code – subagents, skills, containerization, &c – that I'd have to (have Opus...) reimplement.
one feels the llm wow moment whenever what they do on an area has been surpassed by an llm. newer versions of llms are probably trained by the feedback from developer code agent sessions; so this is probably why pro developers started to feel "wow" recently.
the real challenge will be in the frontier of the human knowledge and whether llms will be able to advance things forward or not.
ps1; i'm using 5.3/o4.6/k2.5/m2.5/glm5 and others daily for development - so my work has 1.5x intensified - i tackle increasingly harder problems but llms still really fail big in brand new challenges like i fail too. so i'm more alert than ever.
ps2: syntactical autocomplete used to write 80% of my code; now llms replaced autocomplete but at a semanticlevel; i think and LLM implements most of my actions like a cerebellum for muscle coordination; but sometimes teaching me new info from the net.
So apparently according to Axios this blog post has gone "mega viral" and their article concludes by stating affirmatively that the "AI" revolution is here now. It's been shared by a number of normally trusted sources; my sister linked it to me because she saw Medi Hassan share it with a note that it's the most important thing you'll read in like forever.
To me it reads exactly like every other blog post of it's genre. It substitutes subjective personal experience for any kind of externally verifiable fact, makes appeals to anonymous authorities that always seem to support the author's conclusion, uses language designed to induce a sense of fear if not outright panic in the reader, and at no point addresses the fundamental reality of "AI's" catastrophic unprofitability. Not to mention how gross it feels to read the author's slobbering all over Amodei as some kind of model for good corporate behavior.
Fundamentally my real problem with it is that the author believes that if we make LLMs good enough at coding, they will then become capable of doing all other knowledge work to a high enough standard that they will replace human knowledge workers. That is such a breathtaking example of a Leap to Conclusion that if we could harness it's energy we could start sending spaceships directly to other star systems.
It doesn't take much effort to find news about AI (LLMs) successfully being deployed with ROI in healthcare, legal, customer operations, retail, banking accounting/tax and more. I don't think the article needs to worry about Leaping a Conclusion as there is plenty of evidence outside.
50 comments
[ 3.5 ms ] story [ 94.4 ms ] threadAnd best of all, when it messes up, it doesn't get sued!
You do.
https://xeiaso.net/blog/2026/markdownlang/
The truth of this seems much more banal. Computing has become a major drag. There have been tens of thousands of libraries that reinvent the wheel. Every operating system has become a toy. All major language systems have an absurd learning curve. Each important application is fortified by a giant corporation. Social media is self-important pop babble.
LLMs are surprisingly good at dealing with complex systems. I can fire one up and ask, for example, why this Swift code is not compiling. But why doesn’t my Swift editor explain that problem? Why is it a confusing question at all? The entire system was built from the ground up at enormous expense. Why do I seek outside help?
Our computing is full of whizzy animations and pointless Victorian ironmongery. All meaningless. AI is medicine, not the cure.
They did it first because doing it first was easier. There are tons of examples around and code can be verified to work.
We arent innovating in other areas that might soften the blow. We dont have good support systems, social security, healthcare, or even demands in other areas. How many engineers are going to be plumbers and construction workers?
Jokes aside it should be noted that the author is a founder and ceo of an AI company, not to mention an active investor in the sector. (All disclosed in his "about" page)
Was there something specific in the article you found unconvincing, or that directly counters an experience you've had with AI?
Covid comparison is apt. I remember being insanely scared in Jan 2020 when those videos of Chinese people dropping dead were coming out (and being shamed by most of my peers etc). Few months later it was starting to become obvious it was really only a major risk if you were old or infirm, but the rest of the world had took awhile to catch up.
AI’s big and gonna change stuff - and like COVID probably for the worse - but we’re in a poorly understood hype cycle right now.
Some people really need to be told not to believe everything they see on the media.
Yep! They still remain stupid (in the intelligence sense, not the pejorative sense) tools which have no practical value to anyone with decent skills. But the people who have a financial interest in hype are still trying to convince everyone "it's totally different now bro, use the latest model bro". It's so tiresome.
4 months ago, I tried to build an application mostly vibe-coded. I got impressively far for what I thought was possible, but it bogged down. This past weekend, my friend had OpenClaw build an application of similar complexity in a weekend. The difference is vast.
At work, I wouldn't say I'm one-shotting tasks, but the first shot is doing what used to be a week's work in about an hour, and then the next few hours are polish. Most of the delay in the polish phase is due to the speed of the tooling (e.g. feature branch environment spin up and CI) and the human review at the end of the process.
The side effects people report of lower quality code hitting review are real, but I think that is a matter of training, process and work harness. I see no reason that won't significantly improve.
As I said in another thread a couple days ago, AI is the first technology where everyone is literally having a different experience. Even within my company, there are divergent experiences. But I think we're in world where very soon, companies will be demanding their engineering departments converge to the lived experience of the people who are seeing something like the author. And if they can find people who can actuate that reality, the folks who can't are going to see their options contract precipitously.
Interesting. So you regularly make new apps in 1 hour each.
How is that the same as...writing a book? Did you mean write several short stories? Or are we talking non-fiction?
Yesterday I had to explain to Opus what the color white is and what "bottom right" means after it declared problems fixed, repeatedly, that a literal preschooler would have been able to tell were absolutely unchanged from the original problem description.
I am still waiting for this world of redundant programmers I've been hearing about for years.
I'm not convinced this person knows what they're talking about.
I hope he has a good hosting plan.
"The future is being shaped by a remarkably small number of people".
That is a lot of power in the hands of a few people. Probably nothing to worry about. Power is hardly ever abused...
Does using "@dang" work to get attention to this?
One data point for this thread: the jump from Opus 4.5 to 4.6 is not linear. The minor version number is misleading. In my daily work the capability difference is the largest single-model jump I've experienced, and I don't say that casually — I spent my career making precision measurements.
I keep telling myself I should systematically evaluate GPT-5.3 Codex and the other frontier models. But Opus is so productive now that I can't justify the time. That velocity of entrenchment is itself a signal, and I think it quietly supports the author's thesis.
I'm not a doomer — I'm an optimist about what prepared individuals and communities can do with this. But I shared this article with family and walked them through it in detail before I ever saw it on HN. That should tell you something about where I think we are.
the real challenge will be in the frontier of the human knowledge and whether llms will be able to advance things forward or not.
ps1; i'm using 5.3/o4.6/k2.5/m2.5/glm5 and others daily for development - so my work has 1.5x intensified - i tackle increasingly harder problems but llms still really fail big in brand new challenges like i fail too. so i'm more alert than ever.
ps2: syntactical autocomplete used to write 80% of my code; now llms replaced autocomplete but at a semanticlevel; i think and LLM implements most of my actions like a cerebellum for muscle coordination; but sometimes teaching me new info from the net.
To me it reads exactly like every other blog post of it's genre. It substitutes subjective personal experience for any kind of externally verifiable fact, makes appeals to anonymous authorities that always seem to support the author's conclusion, uses language designed to induce a sense of fear if not outright panic in the reader, and at no point addresses the fundamental reality of "AI's" catastrophic unprofitability. Not to mention how gross it feels to read the author's slobbering all over Amodei as some kind of model for good corporate behavior.
Fundamentally my real problem with it is that the author believes that if we make LLMs good enough at coding, they will then become capable of doing all other knowledge work to a high enough standard that they will replace human knowledge workers. That is such a breathtaking example of a Leap to Conclusion that if we could harness it's energy we could start sending spaceships directly to other star systems.