AI is great at following patterns. AI is great at writing tests. AI is great at doing work that humans would consider drudgery and then applying "enough" judgement to it.
The problem is that most software teams and most engineers are very, very bad at the thing that AI needs to succeed: technical writing. Documenting the patterns, the rules, the heuristics, the SOP, the judgement; developers could barely be bothered to write comments before (which in the AI age are free, infrastructure-less memory that is always "line-of-sight" for the agent and can be used to enrich with keywords).
It's not that AI is the problem, it's that we humans are good at resolving these constraints in the real world at a human pace. AI makes that pace impossible for a human to keep up as a trailing activity. Rather teams need to invest more in technical writing and producing the corpus that AI relies on to learn the operational constraints and rules.
To talk abstractly about all software, you'd also need to analyze all the software that stays working. This article is just cherry picking a few recent incidents and try's to make some conclusions without really looking at causes and whether that can be generalized to "Software".
Software has always been broken at a systemic level. It’s a complicated continuously changing web of abstractions and abstraction breaking hacks attempting to survive prolonged contact with reality.
I do embedded. I’ve always marveled at how ludicrous that “real software” is. I have truths. I have registers that will never be or act differently, I can make pretty damn good assumptions that no one will put this code somewhere else’s etc.
In the past few years though, it’s been creeping into my domain too.
>I have registers that will never be or act differently
Even this needs a qualifier. Radiation exposure can cause bit flips. Certain aerospace applications must address this, and even some earth-bound applications at high altitude.
This feels a little hyperbolic, the grounding was a result of a cable swap and the ssh cert incident they cite was so human caused it’s not even funny. I think things are a little more shaky, but not massively so. I had the town I was living in during college lose its internet and phone lines due to a cable being cut on a mountain pass and it was fine. If your business is sensitive to an email outage, prioritize high reliability over integration advantages that outlook offers. The php2shell vuln is bad, but we also had log4j years ago.
There have been parts of the software industry that have been mission critical from day one and have built processes, languages, and tools to make sure things are 100% reliable. Building software that way is slower and more expensive and in the past has been an overkill for most systems.
The problem software engineers are facing is that now that software is ubiquitous, almost everything needs to be 100% reliable (or close to it). This is coming into conflict with the move fast and break stuff everything needs to be done this quarter at minimal cost mentality.
This narrative isn't consistent with reality. Cloud computing is much older than 10 years ago. We were doing production grade cloud computing over 16 years ago. The OSS revolution dates to the late 90's. The author seems to think that SWE was in a golden age 10 years ago. Sorry, but it's about the same. We have the same problems in SWE as we had 40 years ago. The scale and ambition of the problems we are solving have increased as compute capabilities have increased.
There are some decent points, but this is collating a lot of separate incidents with wildly different causes that just happened to occur in a close timeframe.
I don't even know if there are decent points. This flails around in a way that's just...not super useful to spend the time trying to digest.
Some things happened. AI is also happening, therefore AI is to blame (in the way that some people really believe a problem isn't a problem unless you test for it) and here's a lesson about organizational coherence or something.
These essays are the real AI slop of 2026. Not authored by AI, but always trying to get attention by tangentially being related to AI and getting the anxious upvotes of the "can I turn back the clock?" crowd. And there's a humorous commonality to all of these pieces that they all start with some sort of "don't think I'm not with it. I'm with it! I hand slung an LLM off the shoulder of Orion. I watched C-beams glitter in the dark near the Tannhäuser Gate." call to authority, then pivoting to the "everything has gone to shit" spiel.
We've always had a tenuously connected, barely operated network of systems. We've always had bugs and engineers doing shitty things. Every organization goes through periods of upheaval where all of the institutional knowledge is invalidated or discarded regardless.
My impression is it's always been a mixed bag. The level of bugginess has been relatively constant. In the meantime, scope and capabilities have gone way up. I hate Teams and Outlook and the virus scanner on my company's laptop though. Those are keeping me from doing my work. Windows + anti-virus makes file access and process spawning 10x to 100x slower compared to a Linux machine.
> You can find programmers who know how to manage the complexity, but they want to be paid lots of money.
This wasn't always the case, by the way. It started with the "dot-com" mania and VC investment in the late 1990s. Before that, a programming job paid decently well but it was on the level of other cost-center professional jobs such as accounting. A comfortable living, but nothing extravagant.
Most folks here didn't even know, but a Bitrise security outage hobbled our entire CI this weekend. Friday to Sunday night.
The security issue motivated all Bitrise customers rotating their secrets. this was a big deal for mobile deployments but didn't get much attention on HN..
> About ten years ago, two small revolutions happened in our industry ... And then, we had the Open Source boom.
This seems a bit myopic. Or naive. Open Source "happened" a lot longer than 10 years ago. Microsoft didn't shit itself about Linux just 10 bare years ago. And it's not like prior to 10 years ago, the narrow focus of npm world wasn't cobbling web sites together with bits and pieces garnered from wherever.
That said...
> But AI is both a cause and a symptom
This I agree with 100%. The rise of AI and necessity of AI to continue to make progress is a massive indictment against an industry that prioritized more features and uptime over solid first principles for years, taking out promisory note after note in complexity, in hopes that the stuff papered over would quietly just fade away.
I agree. In my life time there hasn't been a minute of software reliability. There's nothing special about this year or even this month.
I remember the first time I installed Linux in the early 2000s and was and owned within 5 minutes of putting it on the Internet. Desktop computers of any OS crashed regularly. Servers crashed and needed total rebuilds.
If anything software is way more reliable than it appears because the scale of computers and software is thousands of times larger than it was 20 years ago. More things can fail because there are just more things.
People keep believing that the big LLM players with somehow disprove Rice's Theorem. "Somehow we'll be able to get AI to check if a program does X or Y." "We'll be able to have the machines verify things automatically." But this is is of course as hopeless as solving the Halting Problem. Theoretically, we know this is impossible.
Empirically, we can observe the breaking effects of this broken thinking. Code quality goes down, and apps go from being 99% to 95% reliable. But that's not a big deal, they say, that's good enough... But when the quality of all the building blocks further down get degraded in the same way, you can't keep track of which parts are broken, and which parts are dependable, and the whole thing comes crumbling down.
People keep saying, "I can build an app with AI, therefore we can build everything with AI." This article talks about some potential anecdotes that could help to show that doesn't work.
Rice's theorem is across arbitrary turing machines. We intentionally limit the kinds of programs we emit with type systems, knowing they will prevent us from writing some programs, but we'll be able to analyze the ones we can write much better.
We prove properties of programs all the time, that's what a type system is for.
Can we prove all properties about our program that we're interested in? Maybe, maybe not. But Rice's theorem is not the tool for that argument.
Are vibe-coded apps being being written in Rocq programs that perfectly match dependent types to every possible behavior? Is anyone anywhere writing software with type theory at the level where it perfectly limits things and saves our programs from semantic mistakes?
When I first started my career we used to ship like once a month and that was considered good, everything was extremely convoluted and worse than it is today, whole segments of the industry didn't even use Version Control, half the government websites you used stored passwords in the clear and emailed them to you for resets, e-commerce was a cgi script that sent an email for fulfilment and it took weeks to get your shipment, sites went down at night and weekends for maintenance regularly, servers fell over with traffic that would run on a raspberry pi now, but for some reason everyone keeps saying how bad things are now. They're delusional or don't actually remember the past.
It’s basically free to black/grey/whitehat penetration test anything and everything without having to even know what ghidra or metasploit so everything gets penetrated
It's because amateurism has been hailed as the second coming of Christ.
Leftpad, mentioned in TFA, is a case in point: not because of leftpad itself but because of all those defending leftpadism.
"People can't write a proper leftpad".
Yeah, well... If a dev cannot write a proper leftpad, then it should never have been a dev in the first place.
Someone who cannot write leftpad is not qualified to use leftpad as a dependency (and, no, leftpad is neither the Linux kernel nor a cryptographic libray: it's fucking leftpad, which is the entire point).
We've had "devs" whose only qualification was writing glorified "punch the monkey" ads in JavaScript: the JavaScript surveillance/ad complex. Devs who only ever knew the mono-threaded JavaScript ecosystem and suffered from heavy leftpadism.
And those very same surveillance/ad complex bros are now generating countless crap with LLMs, in languages they don't understand.
They're just mediocre.
That's what happened: the veneration of mediocrity.
It doesn't have to be this way.
AI is great at following patterns. AI is great at writing tests. AI is great at doing work that humans would consider drudgery and then applying "enough" judgement to it.
The problem is that most software teams and most engineers are very, very bad at the thing that AI needs to succeed: technical writing. Documenting the patterns, the rules, the heuristics, the SOP, the judgement; developers could barely be bothered to write comments before (which in the AI age are free, infrastructure-less memory that is always "line-of-sight" for the agent and can be used to enrich with keywords).
It's not that AI is the problem, it's that we humans are good at resolving these constraints in the real world at a human pace. AI makes that pace impossible for a human to keep up as a trailing activity. Rather teams need to invest more in technical writing and producing the corpus that AI relies on to learn the operational constraints and rules.
To talk abstractly about all software, you'd also need to analyze all the software that stays working. This article is just cherry picking a few recent incidents and try's to make some conclusions without really looking at causes and whether that can be generalized to "Software".
Software has always been broken at a systemic level. It’s a complicated continuously changing web of abstractions and abstraction breaking hacks attempting to survive prolonged contact with reality.
I do embedded. I’ve always marveled at how ludicrous that “real software” is. I have truths. I have registers that will never be or act differently, I can make pretty damn good assumptions that no one will put this code somewhere else’s etc.
In the past few years though, it’s been creeping into my domain too.
>I have registers that will never be or act differently
Even this needs a qualifier. Radiation exposure can cause bit flips. Certain aerospace applications must address this, and even some earth-bound applications at high altitude.
This feels a little hyperbolic, the grounding was a result of a cable swap and the ssh cert incident they cite was so human caused it’s not even funny. I think things are a little more shaky, but not massively so. I had the town I was living in during college lose its internet and phone lines due to a cable being cut on a mountain pass and it was fine. If your business is sensitive to an email outage, prioritize high reliability over integration advantages that outlook offers. The php2shell vuln is bad, but we also had log4j years ago.
There have been parts of the software industry that have been mission critical from day one and have built processes, languages, and tools to make sure things are 100% reliable. Building software that way is slower and more expensive and in the past has been an overkill for most systems.
The problem software engineers are facing is that now that software is ubiquitous, almost everything needs to be 100% reliable (or close to it). This is coming into conflict with the move fast and break stuff everything needs to be done this quarter at minimal cost mentality.
Hence why we are at the breaking point.
This narrative isn't consistent with reality. Cloud computing is much older than 10 years ago. We were doing production grade cloud computing over 16 years ago. The OSS revolution dates to the late 90's. The author seems to think that SWE was in a golden age 10 years ago. Sorry, but it's about the same. We have the same problems in SWE as we had 40 years ago. The scale and ambition of the problems we are solving have increased as compute capabilities have increased.
There are some decent points, but this is collating a lot of separate incidents with wildly different causes that just happened to occur in a close timeframe.
Let's see how many more incidents will just happen to occur in proximity to those events.
I don't even know if there are decent points. This flails around in a way that's just...not super useful to spend the time trying to digest.
Some things happened. AI is also happening, therefore AI is to blame (in the way that some people really believe a problem isn't a problem unless you test for it) and here's a lesson about organizational coherence or something.
These essays are the real AI slop of 2026. Not authored by AI, but always trying to get attention by tangentially being related to AI and getting the anxious upvotes of the "can I turn back the clock?" crowd. And there's a humorous commonality to all of these pieces that they all start with some sort of "don't think I'm not with it. I'm with it! I hand slung an LLM off the shoulder of Orion. I watched C-beams glitter in the dark near the Tannhäuser Gate." call to authority, then pivoting to the "everything has gone to shit" spiel.
We've always had a tenuously connected, barely operated network of systems. We've always had bugs and engineers doing shitty things. Every organization goes through periods of upheaval where all of the institutional knowledge is invalidated or discarded regardless.
None of this is new.
My impression is it's always been a mixed bag. The level of bugginess has been relatively constant. In the meantime, scope and capabilities have gone way up. I hate Teams and Outlook and the virus scanner on my company's laptop though. Those are keeping me from doing my work. Windows + anti-virus makes file access and process spawning 10x to 100x slower compared to a Linux machine.
> You can find programmers who know how to manage the complexity, but they want to be paid lots of money.
This wasn't always the case, by the way. It started with the "dot-com" mania and VC investment in the late 1990s. Before that, a programming job paid decently well but it was on the level of other cost-center professional jobs such as accounting. A comfortable living, but nothing extravagant.
Most folks here didn't even know, but a Bitrise security outage hobbled our entire CI this weekend. Friday to Sunday night.
The security issue motivated all Bitrise customers rotating their secrets. this was a big deal for mobile deployments but didn't get much attention on HN..
> About ten years ago, two small revolutions happened in our industry ... And then, we had the Open Source boom.
This seems a bit myopic. Or naive. Open Source "happened" a lot longer than 10 years ago. Microsoft didn't shit itself about Linux just 10 bare years ago. And it's not like prior to 10 years ago, the narrow focus of npm world wasn't cobbling web sites together with bits and pieces garnered from wherever.
That said...
> But AI is both a cause and a symptom
This I agree with 100%. The rise of AI and necessity of AI to continue to make progress is a massive indictment against an industry that prioritized more features and uptime over solid first principles for years, taking out promisory note after note in complexity, in hopes that the stuff papered over would quietly just fade away.
He obviously is preparing another doomsday conference talk. Nothing interesting to see here.
On the other hand software is improving dramatically at a systemic level, even without safe system languages arriving on the horizon.
This reads like nostalgia way more than reality
I agree. In my life time there hasn't been a minute of software reliability. There's nothing special about this year or even this month.
I remember the first time I installed Linux in the early 2000s and was and owned within 5 minutes of putting it on the Internet. Desktop computers of any OS crashed regularly. Servers crashed and needed total rebuilds.
If anything software is way more reliable than it appears because the scale of computers and software is thousands of times larger than it was 20 years ago. More things can fail because there are just more things.
People keep believing that the big LLM players with somehow disprove Rice's Theorem. "Somehow we'll be able to get AI to check if a program does X or Y." "We'll be able to have the machines verify things automatically." But this is is of course as hopeless as solving the Halting Problem. Theoretically, we know this is impossible.
Empirically, we can observe the breaking effects of this broken thinking. Code quality goes down, and apps go from being 99% to 95% reliable. But that's not a big deal, they say, that's good enough... But when the quality of all the building blocks further down get degraded in the same way, you can't keep track of which parts are broken, and which parts are dependable, and the whole thing comes crumbling down.
People keep saying, "I can build an app with AI, therefore we can build everything with AI." This article talks about some potential anecdotes that could help to show that doesn't work.
Rice's theorem is across arbitrary turing machines. We intentionally limit the kinds of programs we emit with type systems, knowing they will prevent us from writing some programs, but we'll be able to analyze the ones we can write much better.
We prove properties of programs all the time, that's what a type system is for.
Can we prove all properties about our program that we're interested in? Maybe, maybe not. But Rice's theorem is not the tool for that argument.
Are vibe-coded apps being being written in Rocq programs that perfectly match dependent types to every possible behavior? Is anyone anywhere writing software with type theory at the level where it perfectly limits things and saves our programs from semantic mistakes?
There were package managers long before npm. Maven, rubygems, etc...
Apt and yum.
It was a very long time ago indeed that you needed to manually edit popular programs to build in your environment.
When I first started my career we used to ship like once a month and that was considered good, everything was extremely convoluted and worse than it is today, whole segments of the industry didn't even use Version Control, half the government websites you used stored passwords in the clear and emailed them to you for resets, e-commerce was a cgi script that sent an email for fulfilment and it took weeks to get your shipment, sites went down at night and weekends for maintenance regularly, servers fell over with traffic that would run on a raspberry pi now, but for some reason everyone keeps saying how bad things are now. They're delusional or don't actually remember the past.
Software was mostly always broken but it was hidden and invisible, now it’s visible because the cost of making it visible is so low
This is classic surveillance/detection bias: https://healthjournalism.org/glossary-terms/surveillance-bia...
It’s basically free to black/grey/whitehat penetration test anything and everything without having to even know what ghidra or metasploit so everything gets penetrated
Well there was this saying "Do One Thing and Do It Well", seems once again we are seeing the affects of ignoring that.
It's because amateurism has been hailed as the second coming of Christ.
Leftpad, mentioned in TFA, is a case in point: not because of leftpad itself but because of all those defending leftpadism.
"People can't write a proper leftpad".
Yeah, well... If a dev cannot write a proper leftpad, then it should never have been a dev in the first place.
Someone who cannot write leftpad is not qualified to use leftpad as a dependency (and, no, leftpad is neither the Linux kernel nor a cryptographic libray: it's fucking leftpad, which is the entire point).
We've had "devs" whose only qualification was writing glorified "punch the monkey" ads in JavaScript: the JavaScript surveillance/ad complex. Devs who only ever knew the mono-threaded JavaScript ecosystem and suffered from heavy leftpadism.
And those very same surveillance/ad complex bros are now generating countless crap with LLMs, in languages they don't understand.
They're just mediocre.
That's what happened: the veneration of mediocrity.