Depending on what your job is, this is either the adult equivalent of a snow day or some of the greatest stress in a long time.
I have experienced, half a decade ago during some more frequent but less lengthy O365 errors that there are some higher ups who get when this is not something you can fix and some you very much can't. Wishing strength to any "but it's our mail" and "but we bought the Claude" meetings that happen today.
I agree that it's an opportunity to "do it the old way", as a way to preserve the skill. I actually would like to do that regularly, like half a day every week or something.
But I find it difficult to motivate it from the "productivity" angle. Like if I work for a day without an LLM today, it feels like I could go skiing instead, and the LLM will do what I would have done today in 10min tomorrow...
Which is the way it should be if we were to share the fruits of our labour more evenly in society. Of course in practice the lions share of the product gains go to the capital class and the while they suck the surplus employees and the remaining workers have to carry the burden of supporting the ones now sidelined.
just want to remind HN that it was as little as a year when I mentioned I used LLMs to do all my coding, totally hands off and people here found it a curiosity. One year later and its madness to even contemplate coding manually.
Im not complaining Im just pointing out how quickly the overton window is shifting.
If you’re using the latest models( Opus/Fable/Astra/ maybe Sol) and the default harness, along with a little guidance against the worst tendencies like over-writing tests, you really shouldn’t be getting the wrong things built. Like a coworker, you can’t throw it blind into a monster codebase without some guidance, but otherwise the latest models genuinely just work TM.
Testing, testing, and more testing. In the project I'm working on right now, about 2/3 of the code is in one way or another related to testing. As a human coder, I'd never spend that much time on building test harnesses, automating regressions, writing unit tests, etc, but it's cheap when you can have a robot do it.
I am sure it’s all so good that these incidents should not take more than 5 minutes to resolve. Don’t get me wrong, I am one of the people who would look at you suspiciously a year ago and only in the last year “saw the light”. You were ahead. But that doesn’t mean that at some point you will not overshoot. What we witness today we haven’t witnessed in some time. People also made horrible software but they enriched it with hole plugging hacks that made it run, mostly. AI may build better software but who is plugging the holes. At least today no one. In 1 or 5 years maybe AI. In the meantime, we need to plug some holes and they are not as easy to spot when we are not the ones who left them there.
You give it time....the same way Technical debt is code you chose not to clean up....your Vibe Debt for code you never understood in the first place... will come through.
"Overton window"? It's just developers being obtuse, short-sighted and slow.
It was obvious in 2023 and it is obvious now yet we still have people in denial even today. These people need AI to take over the world and threatening them with eternal suffering before they will change their minds even a bit.
May I ask why you care so much? Some people like hand-coding, they are free to do whatever they want, stop being so weirdly angry and controlling about what people do/how they do it.
Well it's getting philosophical. Even before AI, I would argue that the vast majority of software was meaningless. Actually the most successful software on Earth is hurting society and the environment, in exchange for profit for a few humans.
I don't know any software engineer who can say "if I was not doing what I do, society would get a little worse".
I console myself that while the products we make are useless and probably even harmful the activity itself just represents the world Geist familiarizing itself with computation and its possibilities and thus growing richer in the process.
In the worst case.. sometimes you learn what's best by experiencing what's not.
You might be right in the end, but the middle of a massive outage probably at least somewhat exacerbated by LLMs, being debugged by (presumably) human software engineers is kind of a weird time to make this point
Seriously, if the LLM was down and a dev submitted 10 PRs, I think it would raise some good/bad curiosity from the team with where things stands today.
No, and that's also not necessary. We are reliant on our systems. The correct response is to harden our systems, not to foster some illusion of independence.
We can of course do that privately, for just the heck of it, but it's not the right response for anything at societal level. We simply need to work harder for more of these 9s
What's wrong with failover? I don't use anything from Anthropic except their API. So when Claude is having issues, I switch to GLM 5.3 or something, doesn't take more than a few seconds.
I realise people on the monthly plans can't do it as easily. But I would say that as far as possible, having a plan B is the foundation of resilience.
I think that's the 6 '9's problem. If you have something with 99% uptime, how much more are you willing to pay for 99.9% uptime, then 99.99% uptime. It's risk vs cost. Many places in the US do the same thing with things like say snow removal. Instead of having expensive systems ready to go for the 1-2 days of year of snow, you just shut things down for a day and slow down a little.
That doesn't work for all systems, and not for all businesses, but I would say the mass majority can 'suffer' 24 hours without email or an LLM without everything going up in flames.
Not hard on monthly plans either. I have ant, OpenAI and cursor (mostly for grok and gemiii). Taxonomy of model steps, each has a pre-tuned failover from a different plan and I just need to flip a switch when a lab goes down. Switch isn’t really required but occasional enough that I want eyes on and the system is only for my personal usage so outage until switch is acceptable…
You’re not wrong, but some companies swing real hard at the biggest players assuming they won’t all be down at the same time. My work has hard dependencies on 365 and Cloud Entra and only allows Claude company wide. Today is going to be a snow day since we couldn’t get into anything yesterday but we could at least get projects done using connections Claude had into various platforms that cached the data in 365.
>harden our systems, not to foster some illusion of independence.
Let me know when the tech industry workers stop doing this and gains some solidarity, and I'll join you in that sentiment.
Until then, I pretty much need to bolster myself on the idea that one day I will need to switch frameworks, IDE's, and even OS's themselves. I'm the ones guiding the tools, not vice versa.
>We simply need to work harder for more of these 9s
The only 100% thing I will guarantee is that these tools will align to the incentives of their leadership over time, not the customers using it. add 10 9's if you want, that just guarantees that humanity is the weakest link.
Reading the site guidelines, so I can be Claude-like and properly adjust my answer:
> Comments should get more thoughtful and substantive, not less, as a topic gets more divisive.
You are also a system. And, if you, as a system, don't run with Claude, you're not sufficiently hardened. People have made all sorts of helpful suggestions, like de-risking with multiple providers or just getting yourself a better Mac to run a local model.
The idea that: "No, no, we must have Claude, b/c it's the same as having a pen and paper!" is woefully naive. It's hard to attack the supply chain of pen and paper if you have a few extra reams and a few extra sticks. It's pretty easy for our geopolitical enemies to attack computers located far away from you, especially those systems are designed like fragile Rube Goldberg machines; furthermore, if we build an entire society (a system) to be productive only when we have LLMs, our issue isn't "insufficient-hardening". It's that the design is wrong.
There is always something that you can do that isn’t gated by the api. It isn’t critical path for literally everything. Go organize your expense reports or something.
What do you think I use to organize my expense reports? If there’s something substantial you can do entirely without tokens that’s probably an opportunity for additional automation. But please have failover across multiple labs!
>> If there’s something substantial you can do entirely without tokens that’s probably an opportunity for additional automation.
Well, I just finished a meeting where we diagrammed out how change authority is to be structured as we split one group into two, which teams will retain what authorities. Zero claude involvement. Ten people in a room. Whiteboard and markers with a "scribe" taking notes. I am now editing old SOPs to reflect the decisions made in the meeting ... not something we would trust to claude or any other AI.
>No, and that's also not necessary. We are reliant on our systems. The correct response is to harden our systems, not to foster some illusion of independence.
I'm not sure if this is your intent, but this paragraph reads like satire given current real world events.
The systems you mention in your second sentence are failing at the "right response" you mention in your third, which dictates that contrary to your first sentence alternative action is necessary.
You cannot harden systems you have no control over. Societal level assurances require law and goverment regulation and enforcement. It is juvenile to pretend otherwise. If we were talking about government services, then as a taxpayer you would have some avenue to enact change.
Your only option when dealing with private companies is to try and become anti-fragile as much as possible at a personal level or at the level of your own company.
Is there not a single SaaS, cloud company or similar that your org depends on? No AWS, no Cloudflare, all internal mail servers, no Docusign, no outside inference, nothing of that sort?
Those things are also still pretty loud. And if you have a large shared office space, it might take hours for it to finish.
At our office we have a cleaning crew and when they come by to vacuum during office hours, everyone in our section takes a 5 minute coffee break and then it's done when we get back.
I can't imagine a construction worker saying, power is out, gotta find a way to be productive, I guess instead of the paint sprayer, nail gun and circular saw, we'll just find other ways to be productive with our hands. The same might apply to law, medicine, etc. I think we're almost there with software.
I understand what you’re trying yo day, but my family ran a construction company for generations (until my father retired) and they had contingencies for every adverse scenario to stay on schedule: Generators, battery-operated tools, hand-operated tools, heaters for pretty much anything you can think of for extreme cold, wind coverings, etc. I don’t remember them ever stopping work except for a trade union strike.
I’m pretty sure doctors don’t abandon their patients during a blackout either, or even a war for that matter.
Mature professions have procedures and backup procedures for pretty much every non-catastrophic situation.
Which begs the question is Software Eng'g/Development a mature profession? We've certainly been trying to especially in the last decade, arguably incentivized by ZIRP and VC money but the last few years?
I'm not just talking about tools either. You mentioned redundancies and contingency plans. Layoffs have stripped workforces to barebones headcounts. How can you have redundancy when one sick man-day reduces your eng'g team by 50%?
>Yes, I believe US hospitals are legally required to maintain and test backup generators in case power goes out.
And even that only does so much, it was 45 years ago but my Grandmother had a stroke on the operating table because the power went out and generators didn't kick on until someone could run and manually start them because the autostart failed.
I may be misunderstanding what you wrote, but I think that says more about your imagination than about software. Yes, medics do improvise a lot to keep patients alive when the power is out, or when essential tools or medicines are not available. "Well, the power failed, so did the sustainable backup, let's everyone go home, today's ICU is a writeoff".
Sure, like if your company’s internet is down, you can start sending invoices to customers by paper mail. Just to prove that you’re not reliant on a technology that’s made work more efficient.
I get what you're saying and I would wholeheartedly agree 5-7 years ago but after hearing Our Lord and Savior Claude Code as the solution to all our technical problems with ever increasing fervor and intensity for the last two years, I'm gonna take the cynical route, maybe fence on office chairs, not because my code is compiling, nor my prompt is running, but Claude is down management, what do you want me to do now, actually think?
From what I hear, the project managers are having a hard time sending emails anyway. We'll call it quits, winkwink.
There's also the option where it doesn't affect you at all, because you do not rely on LLMs—or even just do not rely specifically on Claude—at all, and can easily do your job without ever noticing that it's down.
I suppose I can't bring myself to do the drudgery of coding. I enjoy some of the problem solving but not the workflow frictions, spending 20 minutes due to a stupid typos, syntax fuckery, API lookups, etc.
I've started noticing at $WORK that a lot of systems, procedures, etc. that fall under SDLC or devX are being made ever more onerous or complex because "LLMs can do it."
Some examples include requiring that someone addresses every review comment separately, or that a PR description be formatted a specific way, or that support requests have existing JIRA numbers etc. All of these make the developer experience much worse and would never fly if generating text (that no one reads) wasn't cheap now. A lot of this is cargo culting to make vibe coding feel more serious and entreprise.
The teams that have gone full specs driven development are not AFAICT more productive but good luck getting any change in without 10$ of API usage, even for a one liner. In some cases development is now slower with LLM than it was before!
I've seen every one of the things you listed, sometimes all at the same company. None of them add much in the way of effort/time to do the work.
- JIRA ticket - audit trail
- PR format - easier for people to scan, and to make sure things aren't forgotten (you can even have the template inserted when it's created)
- address review comments separately - not sure what to say on this one because I don't understand how _else_ you would address them - I've always addressed each one individually
The team I'm on prides itself in providing great service to our clients (other devs in the company). This includes responding to support requests even if the client isn't entirely sure how to ask or who to ask.
> not sure what to say on this one because I don't understand how _else_ you would address them - I've always addressed each one individually
What I mean is that it used to be enough for you to address PR tasks by changing the code and pushing the updates and marking the tasks as "resolved". Now in some projects at work you also need to reply to the comment explaining how the change you made addressed the task. The idea obviously is to stop LLM from just ticking tasks without actually doing the work, but it's onerous and based on distrust. Not trusting LLMs is warranted but those rules get applied to everyone.
> The team I'm on prides itself in providing great service to our clients (other devs in the company). This includes responding to support requests even if the client isn't entirely sure how to ask or who to ask.
So in that case, you would create the ticket on their behalf before getting to work, right? Because how would you track what you have been working on, if something has happened already in the past, if there is a pattern in support requests, or how well your team reacts to support requests?
> PR format - easier for people to scan, and to make sure things aren't forgotten (you can even have the template inserted when it's created)
The team I just left had a PR template that was probably 60 bullet points (never bothered to count, just grumbled). Certainly was not easy either to scan or to update manually.
It's asking a tractor why the ground is full of dirt, despite having tractors to move the dirt, and tractors to confirm there is absolutely no dirt left before all the tractors can rest.
The tractor informs me that my tractors don't have permission to clean the dirt, nor see whether there is dirt, nor even distinguish that from "no permission to see dirt".
It found that faulty ACL was a known issue, and is a high priority bug for a team I've never heard of, and thousands of tractors are waiting for this.
One tractor found a webcam that overlooks the dirt patch, which bypasses the ACL, but that is now overloaded. Another tractor has setup multicast streaming to broadcast that camera, with low latency, hardware acceleration, and no single point of failure.
We're back in business.
Bypassing ACLs, multicast streaming, and using image recognition APIs for coding usecases are all against company policy, but nobody wants to pull the plug on any of this.
"Tractor, please add the multicast address to my AGENTS.md, and check my tractors have resumed."
"I've updated your AGENTS.md. The tractors have already cleaned the dirt. Here are images from the camera over the past 5 minutes, correlated with logs of the reporter tractors that reviewed progressed, to demonstrate the reports are now accurate too. I found an integer division bug, now fixed. Shall I schedule a tractor to oversee this every hour, by checking the camera?"
"Yes. And raise a blocking bug to fix all this up when the high priority bug is resolved."
Yep. This is what happens. Once you give up control it's mostly gone. The interesting thing is, if you maintain control, you can always choose to use llms here and there. But expecting humans to dive into 10000 file 40million line code bases branded todo apps isn't going to happen.
10k files, 40m LOC is rookie numbers for a todo app.
I recently asked Claude to edit a few targets of a Makefile. It decided: "OH BOY HERE WE GO LETS FIRE UP ALL THE B200S BABY" and went way deep adding tests...I stopped it when I saw something like: "Adding guards to prevent homebrew injection threat...". It had written an entire fucking build subsystem, probably larger than make. For a 5-line change.
Really makes me wonder how much some programmers really understood programming 5 years ago if this is the reaction to a very recent knowledge center being down.
I suppose the equivalent would be Google search being down 10 years ago and being forced to read man pages again. Annoying but not literally "snow day" levels of lost productivity.
Software documentation was a little different when it went through a publication process as a part of a larger waterfall. I was going to say better but I'm not sure about that part, but it was certainly verbose.
You doing your work manually is like going back to the 90s or further even. It feels like dropping into 6502 manual assembly mode. It's fun but it's not "work" anymore. I honestly think without AI my company is better off if we programmers sat around and talked like humans about business. Coding is for the clankers now.
No not really. Most the companies I know that haven't gone full vibes are doing great.
People can and do still write code. Especially in regulated industries. The internet is just full con artists and hype followers not thinking about the future at all.
but I can say that, anyone experienced will notice how you really can't just "let them loose" on the codebase because they'll lose context, not understand the architecture, not have good vision/ideas about the codebase etc.
so I'm bottlenecked by following the code generation and then having to review and respond
and there's so many times where I go "wait but can't we just do x?" and then the model goes, "oh let me see.. yes we can"
if you were to just use agents w/o looking at code you'd have complete spaghetti in no time
maybe the future is that you embrace the spaghetti and just treat code as a black box where all you care about is the input/output
but my intuition tells me that it's not really viable yet. you'll take many steps forward and many steps back.
If all the professionals leave coding to clankers, and discussions about coding to people who code, I see nothing I don't love about that. Just like the web was more interesting before there was money in it, coding is more interesting without people who can't shut up about their work.
That's exactly my point, programming isn't your IT department, "cool things that can be done with computers" isn't even on speaking terms with it, and the web is not a sector of marketing, either. Bores turn everything they touch boring, and now they're telling us we'll be left behind. And I'm like, YES, finally, can't wait, what can I do to accelerate it?
If you are that reliant on LLMs for your workflow, then how do you not have any redundancy in place? Assuming you've have any kind of basic abstraction in place, swapping in API keys for a different provider should be a simple config change...
> If you are that reliant on LLMs for your workflow, then how do you not have any redundancy in place?
You could ask that question about the MS outage as well and million other things.
I suppose I can't understand because where I work our scale is small enough that we can have an actually resilient infrastructure and can stay in control of almost everything we need, and when you get bigger it's more difficult? I mean, even an outage from our email provider could be worked around in a matter of a couple hours at most.
I'm really not sure it's not just because most managers are clueless and shortsighted though.
> we can have an actually resilient infrastructure and can stay in control of almost everything we need, and when you get bigger it's more difficult?
It's a culture thing, if it leans "engineering-heavy", you usually have contingencies, tests restoring backups and lots of other "goodies" that gets skipped if you don't lean that way. Microsoft and similar organizations been a sales-and-enterprise leaning organization for most of its life, so of course they're not gonna think about what happens when their "always online cloud tooling" goes offline when setting things up, it's much more important to finish things quickly and have a long list of features to be able to present to customers.
Depending on company size and security posture (like mine), we only have one provider claude and the company outright bans all other ones, even local models unless approved via a lengthy process
> Depending on company size and security posture (like mine), we only have one provider claude and the company outright bans all other ones
I get it, we had to sign separate HIPAA BAAs with Anthropic and OpenAI - but we did sign them both, if only to be able to have access to a wider array of models.
There are already entire companies out there without a single human who has any idea how their systems are implemented, let alone being able to manually fix them. This problem will only worsen with time.
I think it should be fine for most people. Hopefully people don't rely on Claude for Ops, and if it's not Ops-related but just regular coding, I can't see why it shouldn't be able to wait a few days. I mean, even if Claude is down for a month, people should have enough competent developers to keep going, just at a slightly slower pace, right?
I really don’t feel comfortable merging a PR not reviewed by both of our AI reviewers. They catch so many bugs no human could. Seriously changed my view of code since we started using it in earnest. Bugs shall not pass. Without AI it feels like the Wild West again.
Ditto. I can do it, and thankfully I’ll say AI mostly catches weird edge cases that often aren’t worth the trouble, but having a 2nd set of eyes has been great for quality.
This mindset of "I don't trust myself to do it" is a bit (sorry) pathetic. I think one should realize that mistakes happen and to a certain extent, though professionals have lower rates of failure, everything is a tradeoff.
Some people have asked me, aren't you afraid you'll make a mistake if you fix your own car? Well, aren't they afraid their mechanic will make a mistake?
I think it's good to realize that AI (famously) also makes mistakes, and not to infantilize yourself to the level of not feeling "comfortable" trusting your own thinking.
Though, I do see younger people today whipping out their phone calculator to "double check" something like a 75% increase on 120.
Apologize for strong words, and sorry, nothing personal, but I find this mindset bizarre and even comical. Will we reach a stage where surgeons on the operating table will second-guess their experience and education and refer to an AI assistant?
Imagine we had an AI system that could monitor 38 different patient signals (from blood pressure electrical activity in the brain) during a surgery and give the surgeon warnings whenever these drifted out of the expected range. It feels plausible to me that using such a system might improve surgery outcomes by, say, 20%.
I would prefer to have a surgeon who used the AI assistant rather than a surgeon who chose not to.
Potentially. Probably not in my lifetime, but it's not impossible given how most sci-fi works depict it. You can argue we had a bit of that paradigm shift already with specialized machinery. I'm sure people can still do hand surgeries, but existing tools and sensors increase the precision.
it is a bit sad comparing that to "I can't check for bugs without an AI", though.
Yes, definitely. If an android nails your procedure while you doing it manually drops it into the low, say, 60s survival rate, what would you expect? I fully expect all human workers becoming dependent very quickly once the clankers get good enough.
Unlike what many programmers like to think, programming is not a skilled work on the level of surgery. And besides, many doctors probably would benefit their patients by having AI assist in their work. This has already been experimentally shown for e.g. medical imaging diagnosis.
> Hopefully people don't rely on Claude for Ops, and if it's not Ops-related but just regular coding, I can't see why it shouldn't be able to wait a few days.
IME if anything, people are relying on LLMs for ops more than for development! Devops convinced management that ops isn't actually all that valuable and that devs can just do it. IaC is usually a mess of boring YAML based syntax that no one in their right mind want to learn anyway. Learning Ansible doesn't feel like a transferable skill unlike learning a programming language or paradigm, and in most teams the quality of the IaC stuff (abstraction, code reuse etc.) never mattered much.
As soon as LLMs became good, generating IaC stuff was like the first thing to go full vibe coding, at least for me. I'm proud of my craft, I don't produce bad code with my name on it, I very carefully review all the code changes etc. Devops YAML stuff? Couldn't care less.
Was viewing it more through the lens of less technical companies, Excel, Powerpoint, the backbone of traditional office work has become very model dependent in some companies I have seen and those are the owners and fellow employees that tend to struggle more in getting that the local IT guy can't fix this themselves vs developers.
Just as you described. Snowday for some...
I was caught in between snowday and stress today with these things.
Got too used for claude to just do most things.
But at the same time was surprised I could still google and resolve things myself when I started thinking again lmao.
The most likely scenario, is their team of high achievers, deployed a massive pile of stinking technical debt, against the advice of the experienced folks who are constantly overridden.
On a normal environment the old Graybeards would fix it, after grumping on being taken away from playing with their latest Emacs config.
But this is the Valley, and we dont do experience only VibeOps. And half reading this post, are wondering if Emacs is a new frontend framework they never heard about....
I have fond memories of the “we’re a tech company, why can’t we get a video conferencing call going” and gently reminding them that reliable, synchronous, multiparty video conferencing was a lot harder than anything this company was trying to do.
We are now in year 3-ish of LLMs going mainstream. Do people immediately forgot writing code manually the moment Claude was out? I understand the LLM acts as a multiplier but does coding immediately stop when Claude is down? My line of business is not in software development so this is a truly genuine question.
As people used to say "No one was ever fired for choosing IBM."
If people lose enough money then they will come up with contingency plans etc. that may even work in the next outage! Absolutely no one will be held responsible though. The kind of management folks that made those decisions will just turn around and work on those contingency plans, and everyone (including themselves) will believe that they had been doing so all along.
At least that's my experience in big corporations.
The odd thing about these outages is that when they become so big that everyone is impacted then the severity almost drops, since everyone is impacted and thus just shrugs their shoulders.
Too big to fail. In the 1970s in Ireland there was a 12 month period in the space of a 5 years, where no banks operated due to a huge strike by the workers. The country just ticked along with everyone extending IOUs to each other for when the banks came back online.
It's just one of those things, you'd think the economy or country would collapse effectively but things just tick on by they way they always did. Imagine the banks in the US closing for 6 months?
Yup, it seems like there are a lot of options, but in reality lots of things have become severly centralized. (There has never been a better time to self-host by the way. My Forgejo instance for example is having practically no outages. For AI, use the open models via one of the competing providers.)
I get the sentiment, but the biggest loser is Anthropic, because they aren't earning, and because this will motivate employers to seek alternatives/callbacks.
Copilot is down too, so it's probably Microsoft engineers trying to vibecode their replacement Azure infrastructure before the current garbage fire finishes burning out.
I absolutely hate it, can't understand how that's all they offer (or Google SSO). It's bad enough when the password option is just really hidden and non-obvious (like Notion iirc), not having it at all is ridiculous.
I don't have that email account on my phone. So to log in, I have to open my laptop, get the 'magic' link, find a way to send it to myself on my phone that doesn't make a request (e.g. for an unfurl) expiring it, and then open it on my phone to log in. So magic.
If you open login form and link on different devices, you'll get code to type, no need to send link.
So basically you start login on phone, keep page open. Open laptop, click on link on laptop and you'll get a code. You type that code on your phone and that's all.
There are a number of clipboard sync apps, but personally I swear by simply having an E2E encrypted chat app on both your desktop and phone that lets you send messages to yourself. Serves both the clipboard usecase and as a general "notes to self" space.
I requested Mythos access and was granted it. But now apparently I need to use Google Sign In on my account instead of email link, as email link is not secure enough. But my email is not google, it's my custom domain. Now I don't even know what to do. I don't want to buy google apps or whatever it's called just to be able to use Google SSO with my custom domain; and I don't want to register completely new account for gmail either. It's a mess.
Usually it is the weakest link in the chain, and you are encouraged to switch to proper 2FA or Passkey instead. So even if the email is compromised, other accounts are not compromised.
Is there indication this is really a separate issue or just Claude not being able to reach Microsoft Entra for whatever periodic requests confirm the user is authenticated (and maybe this page even simply being automatic reflection that those requests are errorring frequently)?
Are people who don't use M365 experiencing issues here?
That seems faster than usual, compare to Outlook (down for a day) and GitHub. It's still embarrassing for a heavily-relied service, but better, and I wonder if Claude (or another LLM) helped diagnose and fix the outage?
I look forward to the PR video “How we used a swarm of agents to fix an outage issue”, right after “How we used a swarm of agents to identify thousands of potential drugs for Alzheimer’s”.
If all the hardware innovation didn't get stifled and the prices didn't explode because of all the circular investment deals, a person who needs to step foot in the frontier model territory with local hardware for ~$10k. You could argue that in that timeline frontier model development could have happened differently, but my point is that if the hardware starts catching up again at some point, people who need it, might be going local.
It’s alarming how much of the project knowledge ends up in agentic workflows… although that information is technically on your computer, how long will it take you to find it and unravel the string?
Time to warm up the GPUs I guess, since I can’t be fucked to do it myself lol. Thank gods for qwen flash next and DeepSeek!
TBH the only reason I’m not working mainly with flash next anyway is that I can only run inference for one agent at a time on my setup.
This stuff is so new and wild west at the moment that in our company everyone is using it completely differently. 20% of the devs have 30x output and are doing everything with Claude, 10% won't touch it, the rest are in some weird in between where they use it for some things and not other things which it is perfect for. We have managers that suggest just going back to doing it by hand when there are outages, which is the most laughable thing in the world to me now. It is incredible to watch this miracle tech hit so fast and have everyone's understanding of it be so different across the board.
And jarring. When interacting with colleagues in teams that I haven't worked with, there's an uncomfortable "where are they on AI...", which will decide whether I have to dial down my usage, or whether we can both let it rip.
it is very funny to me that on a story about a major outage at one of the companies pushing the "software is solved" narrative the suggestion to slow down and do things more carefully is laughable, I think we officially need to retire calling ourselves engineers.
I suspect he already used up all the plan B in his immediate vicinity. I feel for the dude, he clearly let shit go to his head. I knew things weren’t going well for him when he put on the leather jacket. It’s basically putting on the “bad boy” costume. If you haven’t purified your soul before donning the jacket, it will consume you, like the one ring.
I used claude for the first time in months yesterday (have been using codex, ran out of resets this week). I had picked up from general consensus comments that 5.5 had fixed the opus 5 issues, but that's far from the case. Its "personality" is still smartass, know it all, and whiny. It disregards specifications in tickets if it thinks it knows better. It complains about the size of tasks when it reads task descriptions. I've gotten used to codex, which just jumps in and does stuff (which of course carries its own set of issues). I have no love for Altman, but their product is still better at the moment.
The post was about their uptime. How is this relevant.
Even if it vaguely is, why do people think subjective anecdotal thoughts about which model is better are a good discussion topic. If there’s an objective evaluation idea that would be useful.
I run evals for a very specific domain. Opus 5.5 is still the industry leader (we can’t use Fable class yet).
It's relevant because there's only so many "yep it's down" comments that are interesting. They have a technology problem which isn't fixed. They're about to go IPO. Their marketing is clearly still working, but subjective experience with both contradicts the marketing.
Going the other way, Claude duet has just gone into invite beta, people get 3 invites and inviting someone else gives them a month of pro, I saw on Twitter a link every minute about 3 hours ago (can't use because I'm Australian) I wonder if they are just getting DDoSed from their own invite spam?
I've been using Claude Duet for a while, and usage for similar chat inputs seems to be > 10x regular Claude chats. I've never hit a weekly quota, but did on Wednesday. My quota reset on Thursday, and I'm already at ~60% of usage.
Depending on how widely this was deployed, perhaps that's affecting usage too...
And this company, about to start its IPO, wants to make the majority of the world's office workforce dependent on them when they can't even handle their current load.
Lmfao I know this should not bring me joy, but it does. Is this the great filter? AI builds AI. Society becomes completely dependent on AI. It goes down, no one can fix it. Total chaos ensues.
It really is Friday the 13th today with outlook and now Claude both being mostly down. I wonder if this is related though. Either way not going to be a great ad for their IPO.
another day, another anthropic outage. Yes, this is because of Microsoft, but they do seem to have a significantly higher amount service disruptions compared to the other LLM providers.
I hope it stays down for the day so my hack "colleagues" flail like imbeciles for the lulz. And I won't have to review their "work" because literally all they do is prompt dear CLAUDE and have everyone read their slop.
I have experienced, half a decade ago during some more frequent but less lengthy O365 errors that there are some higher ups who get when this is not something you can fix and some you very much can't. Wishing strength to any "but it's our mail" and "but we bought the Claude" meetings that happen today.
But I find it difficult to motivate it from the "productivity" angle. Like if I work for a day without an LLM today, it feels like I could go skiing instead, and the LLM will do what I would have done today in 10min tomorrow...
Im not complaining Im just pointing out how quickly the overton window is shifting.
Obviously that's not the case for everybody. 3 years ago, LLMs were very far from producing the quality of code I wanted in my domain.
It was obvious in 2023 and it is obvious now yet we still have people in denial even today. These people need AI to take over the world and threatening them with eternal suffering before they will change their minds even a bit.
If I wait until tomorrow, though, my company won't see the difference.
All the work most people are doing it just going to be Ctrl+A deleted and replaced with someone else's LLM slop using Opus 8 in an afternoon.
All the shit people are building right now just feel ephemeral, like its going to float away on a breeze.
I don't know any software engineer who can say "if I was not doing what I do, society would get a little worse".
In the worst case.. sometimes you learn what's best by experiencing what's not.
We can of course do that privately, for just the heck of it, but it's not the right response for anything at societal level. We simply need to work harder for more of these 9s
I realise people on the monthly plans can't do it as easily. But I would say that as far as possible, having a plan B is the foundation of resilience.
That doesn't work for all systems, and not for all businesses, but I would say the mass majority can 'suffer' 24 hours without email or an LLM without everything going up in flames.
That's not always easy to do, but with something like inference providers it is, so why not kill two birds with one stone?
Another reason to have good Evals…
Let me know when the tech industry workers stop doing this and gains some solidarity, and I'll join you in that sentiment.
Until then, I pretty much need to bolster myself on the idea that one day I will need to switch frameworks, IDE's, and even OS's themselves. I'm the ones guiding the tools, not vice versa.
>We simply need to work harder for more of these 9s
The only 100% thing I will guarantee is that these tools will align to the incentives of their leadership over time, not the customers using it. add 10 9's if you want, that just guarantees that humanity is the weakest link.
> Comments should get more thoughtful and substantive, not less, as a topic gets more divisive.
You are also a system. And, if you, as a system, don't run with Claude, you're not sufficiently hardened. People have made all sorts of helpful suggestions, like de-risking with multiple providers or just getting yourself a better Mac to run a local model.
The idea that: "No, no, we must have Claude, b/c it's the same as having a pen and paper!" is woefully naive. It's hard to attack the supply chain of pen and paper if you have a few extra reams and a few extra sticks. It's pretty easy for our geopolitical enemies to attack computers located far away from you, especially those systems are designed like fragile Rube Goldberg machines; furthermore, if we build an entire society (a system) to be productive only when we have LLMs, our issue isn't "insufficient-hardening". It's that the design is wrong.
Well, I just finished a meeting where we diagrammed out how change authority is to be structured as we split one group into two, which teams will retain what authorities. Zero claude involvement. Ten people in a room. Whiteboard and markers with a "scribe" taking notes. I am now editing old SOPs to reflect the decisions made in the meeting ... not something we would trust to claude or any other AI.
I'm not sure if this is your intent, but this paragraph reads like satire given current real world events.
The systems you mention in your second sentence are failing at the "right response" you mention in your third, which dictates that contrary to your first sentence alternative action is necessary.
You cannot harden systems you have no control over. Societal level assurances require law and goverment regulation and enforcement. It is juvenile to pretend otherwise. If we were talking about government services, then as a taxpayer you would have some avenue to enact change.
Your only option when dealing with private companies is to try and become anti-fragile as much as possible at a personal level or at the level of your own company.
At our office we have a cleaning crew and when they come by to vacuum during office hours, everyone in our section takes a 5 minute coffee break and then it's done when we get back.
I’m pretty sure doctors don’t abandon their patients during a blackout either, or even a war for that matter.
Mature professions have procedures and backup procedures for pretty much every non-catastrophic situation.
I'm not just talking about tools either. You mentioned redundancies and contingency plans. Layoffs have stripped workforces to barebones headcounts. How can you have redundancy when one sick man-day reduces your eng'g team by 50%?
Of course proof by analogy is silly anyways, my email going down is not nearly as bad as a patient's medical equipment going down.
And even that only does so much, it was 45 years ago but my Grandmother had a stroke on the operating table because the power went out and generators didn't kick on until someone could run and manually start them because the autostart failed.
Or maybe more prescient; machine operators going home on a cold day because they can’t get their loader started.
From what I hear, the project managers are having a hard time sending emails anyway. We'll call it quits, winkwink.
That's a scary thought.
Some examples include requiring that someone addresses every review comment separately, or that a PR description be formatted a specific way, or that support requests have existing JIRA numbers etc. All of these make the developer experience much worse and would never fly if generating text (that no one reads) wasn't cheap now. A lot of this is cargo culting to make vibe coding feel more serious and entreprise.
The teams that have gone full specs driven development are not AFAICT more productive but good luck getting any change in without 10$ of API usage, even for a one liner. In some cases development is now slower with LLM than it was before!
- JIRA ticket - audit trail
- PR format - easier for people to scan, and to make sure things aren't forgotten (you can even have the template inserted when it's created)
- address review comments separately - not sure what to say on this one because I don't understand how _else_ you would address them - I've always addressed each one individually
The team I'm on prides itself in providing great service to our clients (other devs in the company). This includes responding to support requests even if the client isn't entirely sure how to ask or who to ask.
> not sure what to say on this one because I don't understand how _else_ you would address them - I've always addressed each one individually
What I mean is that it used to be enough for you to address PR tasks by changing the code and pushing the updates and marking the tasks as "resolved". Now in some projects at work you also need to reply to the comment explaining how the change you made addressed the task. The idea obviously is to stop LLM from just ticking tasks without actually doing the work, but it's onerous and based on distrust. Not trusting LLMs is warranted but those rules get applied to everyone.
So in that case, you would create the ticket on their behalf before getting to work, right? Because how would you track what you have been working on, if something has happened already in the past, if there is a pattern in support requests, or how well your team reacts to support requests?
The team I just left had a PR template that was probably 60 bullet points (never bothered to count, just grumbled). Certainly was not easy either to scan or to update manually.
Those examples sound like responses to being swamped with LLM PRs:
* "every review comment separately" [so the bot doesn't get a chance to miss it]
* "a PR description be formatted a specific way" [so the bot has guardrails]
* "support requests have existing JIRA numbers etc" [so we can have a chance of understanding the context]
> The teams that have gone full specs driven development are not AFAICT more productive
I suppose time will tell. It's difficult to generalise my experiences beyond my immediate team/company/industry.
> In some cases development is now slower with LLM than it was before!
Yes, pre-LLM I wouldn't sign up for formatted PR descriptions!
Tractor breaks, you complain about shoveling again and someone says you’re in the wrong job.
It's asking a tractor why the ground is full of dirt, despite having tractors to move the dirt, and tractors to confirm there is absolutely no dirt left before all the tractors can rest.
The tractor informs me that my tractors don't have permission to clean the dirt, nor see whether there is dirt, nor even distinguish that from "no permission to see dirt".
It found that faulty ACL was a known issue, and is a high priority bug for a team I've never heard of, and thousands of tractors are waiting for this.
One tractor found a webcam that overlooks the dirt patch, which bypasses the ACL, but that is now overloaded. Another tractor has setup multicast streaming to broadcast that camera, with low latency, hardware acceleration, and no single point of failure.
We're back in business.
Bypassing ACLs, multicast streaming, and using image recognition APIs for coding usecases are all against company policy, but nobody wants to pull the plug on any of this.
"Tractor, please add the multicast address to my AGENTS.md, and check my tractors have resumed."
"I've updated your AGENTS.md. The tractors have already cleaned the dirt. Here are images from the camera over the past 5 minutes, correlated with logs of the reporter tractors that reviewed progressed, to demonstrate the reports are now accurate too. I found an integer division bug, now fixed. Shall I schedule a tractor to oversee this every hour, by checking the camera?"
"Yes. And raise a blocking bug to fix all this up when the high priority bug is resolved."
8 minutes have elapsed.
I recently asked Claude to edit a few targets of a Makefile. It decided: "OH BOY HERE WE GO LETS FIRE UP ALL THE B200S BABY" and went way deep adding tests...I stopped it when I saw something like: "Adding guards to prevent homebrew injection threat...". It had written an entire fucking build subsystem, probably larger than make. For a 5-line change.
I suppose the equivalent would be Google search being down 10 years ago and being forced to read man pages again. Annoying but not literally "snow day" levels of lost productivity.
People can and do still write code. Especially in regulated industries. The internet is just full con artists and hype followers not thinking about the future at all.
but I can say that, anyone experienced will notice how you really can't just "let them loose" on the codebase because they'll lose context, not understand the architecture, not have good vision/ideas about the codebase etc.
so I'm bottlenecked by following the code generation and then having to review and respond
and there's so many times where I go "wait but can't we just do x?" and then the model goes, "oh let me see.. yes we can"
if you were to just use agents w/o looking at code you'd have complete spaghetti in no time
maybe the future is that you embrace the spaghetti and just treat code as a black box where all you care about is the input/output
but my intuition tells me that it's not really viable yet. you'll take many steps forward and many steps back.
Is going back to just a couple of months ago. Jesus.
Most people people here do not code. They wish they do, but they don't.
You could ask that question about the MS outage as well and million other things.
I suppose I can't understand because where I work our scale is small enough that we can have an actually resilient infrastructure and can stay in control of almost everything we need, and when you get bigger it's more difficult? I mean, even an outage from our email provider could be worked around in a matter of a couple hours at most.
I'm really not sure it's not just because most managers are clueless and shortsighted though.
It's a culture thing, if it leans "engineering-heavy", you usually have contingencies, tests restoring backups and lots of other "goodies" that gets skipped if you don't lean that way. Microsoft and similar organizations been a sales-and-enterprise leaning organization for most of its life, so of course they're not gonna think about what happens when their "always online cloud tooling" goes offline when setting things up, it's much more important to finish things quickly and have a long list of features to be able to present to customers.
I get it, we had to sign separate HIPAA BAAs with Anthropic and OpenAI - but we did sign them both, if only to be able to have access to a wider array of models.
I’ve been in this situation before, an external API is basically down and my managers solution is to blame me.
Didn’t last there too long.
Why are you still employed? What do you bring to the table if you’ve off loaded so much to AI that you can no longer review a pull request…?
What more needs to happen before you accept that human review does not catch all the bugs that AI does?
Some people have asked me, aren't you afraid you'll make a mistake if you fix your own car? Well, aren't they afraid their mechanic will make a mistake?
I think it's good to realize that AI (famously) also makes mistakes, and not to infantilize yourself to the level of not feeling "comfortable" trusting your own thinking.
Though, I do see younger people today whipping out their phone calculator to "double check" something like a 75% increase on 120.
Imagine we had an AI system that could monitor 38 different patient signals (from blood pressure electrical activity in the brain) during a surgery and give the surgeon warnings whenever these drifted out of the expected range. It feels plausible to me that using such a system might improve surgery outcomes by, say, 20%.
I would prefer to have a surgeon who used the AI assistant rather than a surgeon who chose not to.
Claude is going to write a banger medical drama about this
Amusingly, I did the same and it told me there's no point in taking more than a single dose in my uncomplicated case (prophylactically).
it is a bit sad comparing that to "I can't check for bugs without an AI", though.
IME if anything, people are relying on LLMs for ops more than for development! Devops convinced management that ops isn't actually all that valuable and that devs can just do it. IaC is usually a mess of boring YAML based syntax that no one in their right mind want to learn anyway. Learning Ansible doesn't feel like a transferable skill unlike learning a programming language or paradigm, and in most teams the quality of the IaC stuff (abstraction, code reuse etc.) never mattered much.
As soon as LLMs became good, generating IaC stuff was like the first thing to go full vibe coding, at least for me. I'm proud of my craft, I don't produce bad code with my name on it, I very carefully review all the code changes etc. Devops YAML stuff? Couldn't care less.
But at the same time was surprised I could still google and resolve things myself when I started thinking again lmao.
On a normal environment the old Graybeards would fix it, after grumping on being taken away from playing with their latest Emacs config.
But this is the Valley, and we dont do experience only VibeOps. And half reading this post, are wondering if Emacs is a new frontend framework they never heard about....
My soul just dry-heaved from laughter. Nice.
O364...
If people lose enough money then they will come up with contingency plans etc. that may even work in the next outage! Absolutely no one will be held responsible though. The kind of management folks that made those decisions will just turn around and work on those contingency plans, and everyone (including themselves) will believe that they had been doing so all along.
At least that's my experience in big corporations.
yeah, couldn't be me
It's just one of those things, you'd think the economy or country would collapse effectively but things just tick on by they way they always did. Imagine the banks in the US closing for 6 months?
https://en.wikipedia.org/wiki/Irish_bank_strikes_(1966%E2%80...
Their moat isn't large.
OAuth request failed Temporarily unable to authenticate. Please retry.
Can this be fall-out from the Microsoft outage?
94 points by cloudbreaker 22 hours from now | flag | hide | 64 comments
Copilot is down too, so it's probably Microsoft engineers trying to vibecode their replacement Azure infrastructure before the current garbage fire finishes burning out.
I don't have that email account on my phone. So to log in, I have to open my laptop, get the 'magic' link, find a way to send it to myself on my phone that doesn't make a request (e.g. for an unfurl) expiring it, and then open it on my phone to log in. So magic.
So basically you start login on phone, keep page open. Open laptop, click on link on laptop and you'll get a code. You type that code on your phone and that's all.
Solving problems that don't exist, and creating problems that don't need to.
They just outsource the security into your email provider.
Are people who don't use M365 experiencing issues here?
It seems resolved now.
That seems faster than usual, compare to Outlook (down for a day) and GitHub. It's still embarrassing for a heavily-relied service, but better, and I wonder if Claude (or another LLM) helped diagnose and fix the outage?
This allows their architecture to be simpler, and scale more easily.
GitHub meanwhile is highly relational, across repos and users, and provides real time updates over all this. Outlook is similar.
I guess anecdotal but people nowadays are incredibly addicted to LLMs, to the point where they cant or dont want to work when there's an outage
Time to warm up the GPUs I guess, since I can’t be fucked to do it myself lol. Thank gods for qwen flash next and DeepSeek!
TBH the only reason I’m not working mainly with flash next anyway is that I can only run inference for one agent at a time on my setup.
3 minutes later... I can reconnect, and the "crashed" instance woke up.
And jarring. When interacting with colleagues in teams that I haven't worked with, there's an uncomfortable "where are they on AI...", which will decide whether I have to dial down my usage, or whether we can both let it rip.
In mere months.
currently running a Claude model through github copilot CLI. Pretty odd given the two ongoing outages.
Even if it vaguely is, why do people think subjective anecdotal thoughts about which model is better are a good discussion topic. If there’s an objective evaluation idea that would be useful.
I run evals for a very specific domain. Opus 5.5 is still the industry leader (we can’t use Fable class yet).
It's relevant because there's only so many "yep it's down" comments that are interesting. They have a technology problem which isn't fixed. They're about to go IPO. Their marketing is clearly still working, but subjective experience with both contradicts the marketing.
> I run evals for a very specific domain.
What do your evals test for?
you'd rather have a subject that does exactly as told, or that does what's right?
https://status.cloud.microsoft/m365/
Office 364.3789278392647
(estimated)
Was pretty nice though, less stress.
Depending on how widely this was deployed, perhaps that's affecting usage too...
Well, it was my employers wish that we devs incorporate more AI, and we did. It's their misfortune now.
(I know CC =/= AI, but put that aside for the sake of the joke)
well, it works on my machine.
try this one /s
Jim Cramer has $MSFT as "DON'T BUY"
And now we have Claude and MS365 down.
6 Hours later after this comment [0] Claude also went down. It's now up to the human engineers (if there are any) to fix this outage.
[0] https://news.ycombinator.com/item?id=50478409