Roman citizen wondering why the aqueduct isn’t working kind of thing.
Also undervalued by society as a whole and politicians especially.
They'd rather allocate funding for a new bridge they can point at and say "I did that" than properly fund the unsexy work of making sure the existing bridges are maintained.
Now apply that across most western democracies, repeat for decades and we are where we are.
The countries doing their vast infrastructure build outs will likely repeat the pattern in a few decades.
When banks started moving to computerized systems, maybe a bank would be unable to process something... but not multiple banks. I can't recall any "outages" that affected banks nationwide for any period of time. Walk into your local bank and find out they can't do something because their big system at the mothership is down? Perhaps.
Consolidation likely plays a part in this as well. I can't speak to the FAA/ATC systems 50 years ago, I have no recollection of what they were like then, but the only nationwide system I can think of around that era was Ma Bell. And I don't honestly recall any nationwide outages back then. Regional, sure. You might have a problem with part of the country getting calls if there was some kind of emergency... (e.g., everybody trying to call their grandmas in Kansas after reports of a tornado) but I don't remember there being any nation-wide problems like that.
We had to count on actual snow for snowdays. :-)
There was this one in 1990 [1], but it's notable because the system was otherwise very reliable. Also, this report says only 50% of call attempts failed, presumably most people would try their call a few times and likely get through.
[1] https://users.csc.calpoly.edu/~jdalbey/SWE/Papers/att_collap...
In 2006, Amazon had just launched this new storage service called "S3." It would have struck people in 2006 as bizarre to think that AWS (and Cloudflare, and GCP, and friends) would one day be so crucial to the internet that a bug there could take down dozens or hundreds of companies.
Every service except digital lending/materials continued to function, and there were fallback processes for recording everything that happened so the information could be added back into the digital system once the tech was fixed/power was restored. Specifically designed forms that were easy to use and easy to scan, an exact understanding of what needed to be recorded and why and where it would go later, etc. And a staff of people who were meticulous enough to actually follow through.
When I hear people talk about "we can't do big things anymore" in the US, they are usually referring to mid 20th century stuff a la https://patrickcollison.com/fast, not stuff that we did 10 years ago.
I do not believe the FAA has ever grounded the entire country due to a network upgrade going bad. So that is possibly an example of a hard thing we could do a decade ago, that we can't do now.
Not the things that are breaking, no. The FAA was doing ATC a decade ago. We had email and cloud servers a decade ago. We had bank websites a decade ago. The scale of those things today is perhaps larger than a decade ago, but not enormously larger; indeed, the increase in scale in those things is much smaller than the increase in computing power in the world from a decade ago. So these things should have gotten more reliable--but they haven't.
we made these systems simpler to maintain, to reduce the amount of skill we need to pay for to maintain them. And that does reduce the maintenance costs, at the cost of more compute. but who cares about compute costs where compute per unit of money doubles every 18 months, amirite?
unfortunately it also increases the competence required to change these systems without degrading their performance and usability.
since no one wants to pay that extra cost either, we let the systems degrade a little bit at every change instead, hoping they degrade slower than hardware improves.
that bet was bound to lose eventually, because of laws of physics, and because that degradation pattern affects hardware too.
and the double whammy is - with little demand for expertise - the supply is effectively gone too. so now that the walls have closed in, there is no way out.
[1]https://www.fintechfutures.com/baas/bt-fault-hits-uk-banks-d...
[2]https://abcnews.com/US/hackers-block-us-banks-online-service...
If I were you (or was working at somewhere with an outage), even if it wasn't AI, I would still be blaming AI. Because the alternative means you shouldn't have a job in tech.
I was able to log into my bank's website in the late 90s, FWIW. Yet today I can't seem to do that.
But for everything else you need, food water shelter healthcare, you need to submit to europay visa mastercard
They are entirely wrong.
None of these outages, as far as we know, have a common cause. They just happened by coincidence inside the same news cycle.
Nor is there any evidence to blame them on AI, or tech layoffs, or hackers from Iran. It's more likely something boring and technical that doesn't fit into anyone's narrative.
Not sure, I run a simple website for my oss work: https://github.com/mickael-kerjean/filestash yesterday, I investigated a reflection attack that was bringing an entire network down, the ip originated from both china and iran. The customer brought me in there was in panic, he spent 25 years as a msp, but that kind of attack was a first to him. Over the last 2 weeks, I have received about 20 security reports when the usual rate is once every few months, and on top of that literally every single form coming on my website is being abused by spam, the rate of spam message coming through there has 100x increase over the last few weeks
1. Does there seem to be a larger concentration of outages recently? (I would say yes.)
2. Is there actually a larger concentration of outages recently, or does it only seem that way because of an increased focus on this? (I would say that there is really a larger concentration, but I'm less certain about this.)
3. If there is actually a larger concentration of outages recently, is that because of AI? (This is the point at which I become pretty unsure either way.)
4. If it is because of AI, how exactly is AI causing this? (I think more widespread vulnerability discovery / exploitation is more likely than "everyone is vibecoding everything now and that's screwing everything up", but I'm not entirely sure.)
5. Is this just transitional growing pains, or is this a new normal, or is this the beginning phase of a pattern that is only going to accelerate? (I don't know the answer to this one at all! I hope it's the first thing, but I wish I were more confident.)
Either way, now that AI is more capable, you can really mess stuff up at scale.
You saying they *are* entirely wrong is as baseless as their claim. Sure, you can say they don't have a common cause, and I agree with that, but chances of these events with different causes occurring so close together is very low and the chances of a same root cause is probably more plausible.
One way or another, they've tried to avoid addressing tech debt, avoid hiring and retaining good people, and have more or less been kicking the can down the road / running on the edge without knowing it.
Or else they did embrace AI-driven "move fast and break things" approaches, and so they broke things.
They flip-flop between maximizing profits and draining operational divisions of any resources, and hoping to sprinkle a bunch of IBM magic pixie dust all over their infrastructure to solve all their problems without needing to understand any of it. You rarely find one that strives to understand problems deeply and invests in long-term planning and maintenance before a crisis hits.
The result is old systems that eventually break, system upgrades that go bad because nobody understands them, security holes that someone might have hacked into, etc. All the same basic root cause at the top, which is that most of our leadership class is full of incompetent short-sighted narcissists. And it "works" because most of the time it pumps the stock due to lower costs, and we only see the fraction of times that it causes headline-grabbing catastrophes. On a long enough timeline, though, they'll happily drive our entire society off a cliff--and we're getting closer and closer to that (at least politically).
It's your job to provide evidence they are related, and you have nothing except speculation.
It’s actually hilarious that you made the irrefutable claim, and then you are telling me to find proof.
They'd really be the only people with information about the cause of the issue; if they blame an AI-generated code change or the actions of an AI agent, that would be good evidence.
There are many more systems in the world outside of it, even if it is relatively ubiquitous at this point.
"As far as we know" just indicates a lack of data, not evidence that they're not related. This is especially relevant for a political administration whose response to COVID was to bury their head in the sand by hiding data. Also: A public comment of "AI is hacking all our shit" wouldn't exactly calm the public.
It's fine to say, "I'll wait until we know more", but in the meantime AI progress continues and the sheer volume of cyber vulnerabilities it's uncovering is undeniable. It's not surprising that some state-sponsored bad actors would be use this as an opportunity to infiltrate too...
In the absence of data, you are assigning blame to things you already hate, like AI or the administration. Motivated reasoning is highly prone to error, and it's a form of stupid-brained thinking.
Microsoft on the other hand, has proven that it will never be reliable.
You can argue that this is a result of something outside of their control, and you would not be wrong, but if the point here is about centralization rather than simply a particular company's operational procedures, then direct strikes on datacenters should be part of that conversation.
I was an Amazon red badge in the EC2 org, fwiw, and do think AWS has excellent operational culture, far beyond what I saw while working in the M365 org, and beyond what I've seen at other tech companies and hedge funds where I've worked. But that doesn't obviate issues with centralization.
If Bank of America's central datacenter (cloud or on-prem) gets hit by a cluster bomb, taking out banking for ~half the US, you can be damn sure that very smart people are going to be working 24/7 until it's resolved.
If you have all your savings stored with a local credit union, say Seattle Credit Union which has 50,000 customers, and they get hit with a ransomware attack... well, good luck. You (probably) won't lose any actual money thanks to FDIC rules, but it might be weeks or months before their services are back online.
Similar to self-hosting vs cloud, you get a bit of a paradox where it's overall better for society if there's no interdependencies so a SPoF outage doesn't impact everyone, but things are much worse for an individual in the event a localized outage does occur due to economies of scale, or lack thereof in this case.
Its pros/cons are known and gaps can be addressed with 3p software. You just keep on old outdated, but proven version of software and keep the perimeter locked tightly.
its the azure cloud that is the wildcard.
Don't forget that MS has the advantage of non-English languages support that many alternatives lack.
I feel like it is very unknown, there is no documentation and everyone relies on comments on random forums that are now 21 years old which tells to set undocumented registry keys to undocumented values.
It's not just Europe! :(
I prefer to ask a different question. Will all this AI development delivering us amazing new software? Are we just going to get 1000 free clones of Adobe Photoshop or can we expect some really exciting groundbreaking new tools? Perhaps all the really clever use of AI development is in the pursuit of better AI and if that’s the case, why?
As to why, well economy must grow, no matter what.
Unfortunately it will take many more outages for folks to realize that serious engineering talent has great value in the AI age. It's their judgement that helps speeding headfirst into building unreliable unmaintainable and difficult to extend systems.
Some other thoughts:
* Will the Internet become balkanised?
* These smaller networks will require a lot of personal information to even gain access.
* The return of the Sneakernet?
Not the same root cause of course, but there's something in the water
not sure if vibecoding is a cause or consequence, but it’s related, it requires as an axiom to stop caring about internals. And it seems to require disregard for quality too.
i'm all for experimentation and adding new tools when they're shown to actually help, but it seems like large software companies (especially banks and e-commerce) just don't care about quality anymore.
Commonly asserted but not true. Most programmers are not just letting the model take the wheel.
> it'll probably be part of the solution.
LLMs can't solve the slop problem they themselves are creating.
In what way did you interpret vibecoding as taking the wheel away from programmers? It certainly does in writing code. Generally developers are still involved in architecture design and defining requirements to give their agents.
> LLMs can't solve the slop problem they themselves are creating.
Human slop is the real problem that LLMs can and are solving.
All of this is also downstream of what I think is even more important, which is what's better for society in general. Obviously low quality software is not better for society than high quality in a vaccuum, but I also don't think there's a single obvious objective answer for whether the influence of software on society is a net positive or negative today compared to, for example, when computers were all mainframes and all software was produced by people paid to write it. There are a lot of differences in society today compared to back then, and software certainly has some role in how we ended up here, but trying to attribute exactly what and how much is not an exact science.
This is a lot of words to basically say: if you're confident that this is a problem, I think it's important to state exactly why, since I think the interest parts of the discussion are in how you've concluded that.
Maybe it was because I was a lot younger back then and a lot more idealistic, but especially for the ~2005 (i.e. just after the dot-com crash) - 2010 period (just before money would ruin everything in this industry) I can say that the programming bar was placed a lot, a lot higher compared to where I see it placed now. Software wasn't perfect back then, of course, but there was a real push (belief, maybe?) that with enough will and with enough new-knowledge-being-acquired then we (meaning all of us as an industry) will make things a lot better. In part the existence of this very forum is a simptom of that philosophy, what if it (originally, maybe it still is) being written in Arc, or Reddit originally being launched in Lisp.
Vibecoding is an extension of the culture of overrelying on massive external dependency graphs. It's a preference for visual confirmation that something works and a disinterest in how it works. I think it's also related to the tendency of content feeds that are based on likes and swipes instead of searches, instead of specifying what we want, we are reduced to approving or dissaproving (until we are removed from the loop entirely)
Uptime and quality of software is about 10000x better than it was 10 years ago. Things just work all the time now.
Extraordinary claims require extraordinary evidence. You're being the 'trust me bro' in this situation.
The absence of something that once existed is not the same as that thing never having existed. A null entry and a non-existent entry aren't the same.
/s
People "vibing" everything is just another component of this, not the root cause. Using AI tooling is just the new technique for what used to be called phoning it in, or going through the motions, or faking it until you make it.
Unfortunately, it's nearly always been an effective tactic for a selfish few. It's problematic when we don't have an equilibrium that keeps it in the minority...
(The same thing happened with separate company's problems becoming "Big Tech." It didn't help.)
Snarky, but there's a point: Abstraction leads to viewing different things as the same thing. They are - at one level of abstraction. And they aren't at other levels of abstraction. It's not a problem to have both viewpoints.
The trick is that, when someone blends things together, to not lose the fact that they are also separate (at other levels of abstraction).
So I have no problem saying that "optimizing for efficiency and speed" is the common cause of all of these problems, and saying that they all have different (immediate) causes. Both can be true.
I'm not sure "optimizing for speed and efficiency" can count as a common cause, though, because it's not a single event. It's more like a set of events that have something in common.
Now when people have a question, they ask AI, which is wrong a lot of the time.
The main incentive for any nation state to avoid attribution to direct and material harm to many Americans is that they risk awakening American bloodlust. Americans are unhappy about the current Iran situation and wish it didn't start and that it would end already, but sufficient Iranian escalation would turn that around real quick and suddenly there are boots on the ground among other things.
Saying it's not related applies an assumption of what level of specificity you are thinking about. And that's how you quickly de-sync with other people, especially "lay people" who look at you funny when you claim there's just no way these things are related when it's hard to believe that based on experience alone.
As other comments suggest, it probably is related to an extent. Related by complexity of our Internet-based systems, related by age, related maybe even by enshittification of expertise, etc.
The further up the tree you find these relationships, the less specifically actionable they are (you cannot CVE enshittification), and the more they are about "what are the overall social/cultural/human issues we're grappling with in this era?"
Sure, focusing on individual small causes is helpful too, and it's a lot harder to do the higher and higher level system stuff. But, it can be done.
edit, P.S.: It's worth keeping a bare level of suspicions that there could be a common lower level cause, whether through inadvertent dependencies added or through hostile action.
That's the preliminary post mortem from Microsoft.
I don't think we'll ever get anything honest out of the administration for the FAA outage unless Congress really wants to dig into it and make the results public.
Must be the water then
Apparently the FAA outage was caused by the pressure to do things quickly, fueled by the AI urgency.
It feels like all the IT industry is feeling this pressure.
b) there is also a risk of sophistry and language games obscuring common sense truth
In the world where a state actor is engaging in active sabotage as part of a cyberwarfare effort, there is one shared cause.
Which exploits are responsible in each case may differ. Still one cause.
Do I think it's Iran?
No idea.
But the point is, none of us here have any idea at present.
- Massive Microsoft outage - FAA ground stop - Now this outage
Definitely a black swan at least? It’s not even the number of failures, it’s the scale.
I told him it's just like how he keeps telling his patients to watch what they eat, drink and smoke less, and get some exercise, but they don't, and then they have a heart attack and wonder why.
I guess for most “technology” is one big blurry mess.
Wait till they hear about United Airlines.
But I think there's an increasing consensus among "normies" like your coworkers who see these as related. They don't know why but assuring people they are unrelated fails in messaging terms when it feels related.
I think this will be an increasing problem where so much more code is not only written by AI but reviewed by AI. Things will slip through and companies won't be punished for these outages in terms of public perception (because it'll happen to everyone) or by the government. We just don't have serious consequences for corporate malfeasance or gross negligence. Like AFAICT the Sackler family should all die penniless in 6x8' grey prison cells.
Unfortunately we live in a country where the Sacklers get to stay billionaires while 1 year olds are shot dead at the behest of Walmart over a pack of diapers that weren't even stolen... with zero consequences [1].
It feels like this is just the next level of entshittification.
[1]: https://www.nbcnews.com/news/us-news/mississippi-ag-says-pol...
AI is acting as a sudden precipitator for problems that have been latent and growing for at least 15-20 years.
In pursuit of maximally rapid innovation and reduced capital expenditure, the software industry adopted a "move fast and break things" mentality and homogenized code and service infrastructure over a few winning cloud vendors, SaaS providers, and code solutions/frameworks/packages/etc.
The pitch was that you could rely on downstream experts to do better work than you could, that the biggest vendors and publishers had the best experts, that have them invest the capital would look better on your accounts, and that their scale would give you the best value for your investment.
The caveat is that effectively all technology products have serious faults lurking in them, even with the finest expertise working on them (not that most actually had that), and that when everybody homogenizes over the same providers and solution the industry shifts from a bubbling froth of localized bugs to a heavily correlated house of cards where widely correlated failures manifest in widespread collapse and catastrophe.
Whoops.
AI is acting as a breeze against the house of cards, but lots of things could have and would have and did, especially as it continued to inflate year over year.
But you did speak directly to AI, and specifically to advocate for regulation of it, which is why I replied to you. You were being specific and material in a way that the parent commenter was not, and that invited a different response.
You give other people way too much credit.