And we know why. It's because of their hostile attitude towards their users.
No doubt they'll continue operating until there's no more value in selling AI data licenses, then they'll unplug the servers and unregister the business.
That’s why it succeeded. It was not a forum, it was something new. Contribute a new question that was well stated and broadly applicable and get a response. You had to earn your answers.
It wasn’t flawless, but it was good and so good it fed all the AI agents with good materials.
It wasn’t a “help me!!!111!” site, it was an exchange.
I put lots of effort into narrowing my questions, eliminating obvious answers by demonstrating the work I put in, and editted all my questions like an email to my boss. I was almost always rewarded with good answers.
Huh. I had the opposite experience. The better my questions got, the fewer responses I got.
I stopped bothering, after a number of questions went answered only by me (after I found the solution on my own).
It did reduce the insults, but that seemed to be the only gain that I got, for carefully articulating my questions.
LLMs aren't perfect, but they don't imply that I'm being lazy, for not already having the answer, and they do give me an answer, instantly (and, like SO, the answer is not always the best one).
I have engaged with some younger devs who don't even bother turning the thought into a question. They make statements and expect you to do the work of inferring the question. Which I feel like they got from chatbots since it's faster and LLMs don't care if you make them jump through hoops.
> The better my questions got, the fewer responses I got.
There is a sweet spot of providing just enough detail but not too much in asking a good question. And even the best only “bat 300”. I've posted answers to my own questions plenty of times when I got no response.
I hear you on the hostility. Also a lot of people didn't experience how bad things were BEFORE stack overflow. The first time I tried asking a question on IRC I got mocked so bad I logged out and never tried again. Most forums were like that.
I gravitated to Ruby back then because it was one of the only forums people would be patient and empathetic with me.
Hear you on Ye Days of Yore. I never did IRC, but I did try UseNet, and it was basically a public toilet.
Everyone bitches about the Eternal September, but, in my opinion, it actually reduced the hostility and barbarism. Clueless dolts were a welcome break from the malevolent assholes.
I have repeatedly had this experience in trying to craft The perfect question for a consultant who was also a mathematics professor, in finding the essence of the question the answer becomes clear and I never even have to send the email
Yes, that was part of the point. I hope you posted anyway (after checking for a duplicate, and assuming the underlying problem wasn't just a typo), and provided your own answer for it. That is an explicit part of how the site is supposed to work (but an incredibly poorly communicated one).
> Also a lot of people didn't experience how bad things were BEFORE stack overflow.
SO actually considerably improved on that axis over time. It's just that tons of people erroneously think that being expected to meet certain standards in asking a question (in particular, doing the necessary debugging work to locate and reproduce a problem, or the necessary contemplation to properly specify a task), is the same thing as yelling at people for not knowing answers to easy things. That goes both ways: people get upset when their question is closed, and people think they're helping by excluding easily answered questions because they're easily answered. In reality, the main problem is questions that are easy but complex, i.e. which really break down into multiple separate easy questions or require digging through a code dump to find a trivial error.
The “answer my own question” feature was meant for exactly that. Had quite a few like that, and sometimes I still got better answers than I found
60000 rep btw
or just a great form of your question if someone genuinely wanted to answer it, not pump their score, dopamine etc?
And _then_ it became so much more hostile to users - and lost a lot of what made it so wonderful. It became very common that asking a new, good question would be marked as a duplicate because there was something it wasn't a duplicate of.
What amazes me is that an LLM can actually make sense out of all this crappy training data and often produces meaningful answers. (maybe that's a sign that we already have AGI)
I detested it but quickly realized that I needed to put in work and learn and only when I understand can I ask valid questions to real problems…not just asking for someone to spoon feed me an answer.
I really appreciate that experience…
The moderation style in the latter years was very different than at the beginning.
It was always heavily moderated, but after some point it felt like there were more layers of moderation and rules and cultural expectations added every year.
Moderation went from being a background cleanup operation that quietly kept the site in line to being the primary thing I expected to see when following a StackOverflow link from Google. In the later years it felt like every other StackOverflow search result was going to be the question I had followed by a moderator shutting it down for some reason before anyone could answer.
The moderation wore me out and I wasn’t even posting questions, just trying to find answers!
No, it absolutely very much was not. Even granting that you mean actions of the community (what we call "curation") rather than the small and extremely strained moderation team, way too much stuff historically got through, while people yelled at the OP in the comments and argued over downvotes instead of taking the action that actually matters: voting or flagging to close the question. Worse, people would post answers that acknowledge the problem with the question and asking OP in the answer text to fix things.
I know because I was part of that mess back then, because I didn't know better. That, in turn, was mostly because the entire set of ground expectations was poorly communicated, and the community wasn't given the tools to improve that communication (e.g. couldn't edit the help pages, had their feedback about the discoverability of site documentation ignored, etc.). But it was also because Atwood and Spolsky hadn't refined their intuition about why beginners' bad questions are bad. (Hint: it is not because most of their problems are easy to solve, or "beneath" experts.)
> when following a StackOverflow link from Google
Google got worse at search over the years, so you ended up seeing low-quality versions of the question more often. And the site got way too big (there is no reason why there should be more than three times as many publicly visible questions on Stack Overflow, as publicly visible articles on Wikipedia), such that search results could get diluted with irrelevant low-quality questions that happened to have a lot of keyword overlap with what you wanted. And misleading questions didn't get cleaned up, so you would search for X and find a question where the OP hadn't tried to analyze the situation at all and ended up asking about X when the problem was actually Y, and got the question closed as a duplicate of the main Y "canonical". (I know because I used to lose considerable amounts of hair trying to find the X canonical that I knew was there but stubbornly refused to turn up in searches.)
If "moderation" had actually "always" been "heavy" then you wouldn't have had this problem because this stuff would have gotten hidden from public view, one way or another.
> a moderator shutting it down
Again, these actions are overwhelmingly not taken by moderators, and phrasing things this way reflects a refusal to interact with the community on its own terms and understand its policies. Those who hold an outsider mindset are bound to end up feeling like outsiders.
You know that users are going to call anything that happens to their posts on any website “moderation” right, and after decades of that being the norm that’s what the word means
Actually, talking to ChatGpt, I found this article: “Stack Overflow Isn’t Very Welcoming. It’s Time for That to Change.” ( https://stackoverflow.blog/2018/04/26/stack-overflow-isnt-ve...) and that actually shows what the SO admins didn't get (imo obviously).
They thought the problem was people not being nice. But that actual problem was that SO was "gamified" and it's gamified interface made it impossible to win unless someone was ultra-competitive (and I'd those types burned-out over time as well). People not being nice was real but it was side-effect of the situation (questions answered instantly so a would-be question answered had to be constantly racing).
It's easy to imagine how the admins didn't see the situation. Each up of intensity seemed to up the engagement and so seemed a good thing. The question of whether the process was sustainable didn't appear and when there were problems it was attributed to something else.
I think
From my experience in Home Improvement and Bicycles, I almost always get in some form of contention with the mods over the boundary between "broadly applicable" and "too specific". I'd do the esr-minimum[1] of showing my homework by posting pictures of my WIP and citing Youtube videos I'm trying to follow but, apparently, this is already toeing the line. I honestly don't know what the mods expect of a good question there. Of course, there will be an appreciable degree of case-specificity in my questions because if the YT tutorials could already answer my problem, I would have no need to ask for help and advice!
Then we have sites like Literature SE which has been languishing in their sandbox since literally forever. Like most SE sites, the guidelines bid you to "avoid asking for opinions or open-ended discussions" and "be specific" but I find that untenable in a Literature site. Notably, sister site SF&F is comparably thriving but there is a fair amount of open-ended speculation going on in there.
Lastly, IMO, it just suffers because at its core its problem is human and social, not technical. In an effort for better curation we ended up with Computer Science, Theoretical Computer Science, and Computational Science. For some posters, finding the right SE site already introduces a huge wall to overcome for posting. This hampers a lot of the out group: students who might be seeking homework guidance or looking to supplement their own readings and lectures, amateurs who are just curious about something in the field. I've had a question in one of the Science SEs outside my field that was well-received by the community but one mod kept "tagging to delete" (I forgot the exact term) almost every year. This mod action has a prescriptive period and, boy, he kept at it. Honestly, I don't care in the end. I asked the question out of sheer curiosity and got a satisfactory answer. Yet I wonder how exactly they expect to foster a community this way.
If StackOverflow is being considered hostile now it's big ... well, Google the three stages of enshittification I suppose.
To be fair not really but it’s a name I haven’t heard in a long time and all it took was to scroll down to the end of the page to get your answer.
Real 20 year ago internet there.
Really? Because I only ever asked three things, technically two I suppose, and the only significant interaction I got from there was a guy saying they thought the third question was a duplicate but they only didn't mark it that way because they saw it was me asking again.
This dude straight up admitted to be on the hunt for marking things as duplicate and only put brainpower into doing not-that because something felt off in is gut.
Then someone else did.
No, he didn't actually answer my question. No one did.
That's exactly what it turned into, though.
https://meta.stackoverflow.com/questions/320536/what-is-the-...
Right, their decline is because of the thing that they've been accused of since the very beginning, and not the trillion-dollar technology that replaces them. Brilliant take.
Now we have a new guy in town, who offers "robotized copy-paste the solution to YOUR specific programming problem, no reading required! we can even control your computer for you, so you can do something you actually enjoy and stop wasting time on this complicated computery thingy for boring nerds", which serves the aforementioned audience much better.
you sound exactly like the mfs that ruined that forum.
But also, we were not "mods". We were just members of the community who did absolutely nothing to gain power over you except to participate by asking and answering questions. The actual moderators of the site, with a diamond next to their names, were almost literally one-in-a-million users. But it only took 3,000 reputation to cast a close vote, and 15 to flag to close a question (prioritizing it for consideration by the first group).
Posts like yours are made, IMX, because people had a bad experience asking a question. If that describes you, feel free to provide a link to the question and I'll be happy to give a detailed explanation of why things went as they did and what the site wanted you to do better in asking it.
The one thing I really liked was how it tried to find existing questions and answers that were like yours during the question input phase, but that is also why it came off as hostile when you subsequently collected down votes even after passing through this filter
This is what killed SO. People were going to the site to get problems solved, not to add to their list of problems.
"Target audience" connotes people that the site wanted to be there. That's just who showed up, because that's who will always show up basically anywhere that looks like it might be helpful to them, by default.
The biggest suckers were the people who were genuinely putting it effort there. So, in the end, did they make the world a better place? One would like to hope they did, but we are yet to see the long term consequences of the capabilities which were created with the value extracted there.
We weren't going to be a huge contract (~100k/yr) but the tone was very much "conform or fuck off". I think the sticking point was that we were offered 1x permanent careers ad in North America and 2 in Oceana, and we wanted 3 in Oceana instead.
I've had a similar experience the couple of times I've asked or attempted to respond to questions (about software I wrote lol) - grammar and framing seemed more important than technical information/a solution.
N=1 anecdata ofc, but I'm not surprised at their demise
That and [redacted] mods who delete your question “because its a duplicate” but its not, you just assume it is and lock it. Lots of this when .NET Core came out and it drove me up a wall. Wouldnt surprise me if a lot of “hallucination” type answers come from bad Stack Overflow data being interpreted as “fact” in earlier models.
My extensive experience has been that the overwhelming majority of the time that people complain "this isn't a duplicate", it actually quite obviously is. (And there were even more irritating complaints on the meta site at times; all the way up to the level of "you shouldn't ever close duplicates at all because I think you shouldn't", demonstrating no interest in even trying to understand the reasons for doing so.) Of course, people do make mistakes; there was an entire process for fixing this, but most people could not be bothered to engage with it productively (or even civilly). But feel free to give examples of things that you feel were wrongly closed from your own experience.
Q: How do Python's dict preserve insertion order?
A: [duplicate of "Why does a Python dict's keys have a different ordering each run?"]
It was clear that the mod/curator/whatever spent no time attempting to understand what was actually being asked, and linked it to something that was completely incorrect as an answer to the real question.
I saw this constantly, so frequently that I'd be hard pressed to answer with specifics. If you asked me "when did you last eat a fruit bat", I can tell you the month and year because that stood out. If you asked me "when did you last eat spaghetti", I'd have a hard time saying.
Fair, I got my wording wrong, its been a while since I've landed on the site... Sadly, I used to love reading SO for fun and to learn random tidbits.
I've run into way too many "this is a duplicate of" and its referencing something from 2003 in 2025 where the runtime has changed, and a similar looking issue was confused for the actual issue, the resolution in the linked "duplicate" doesn't apply or work. In my opinion, if you're going to call something a duplicate, prove that the solution there works before you lock it down.
You must be new to IT. Software businesses rarely truly die off (unless they're really small one person operations). Even Expertsexchange is still around: https://go.experts-exchange.com/
Uh, no.
StackOverflow is and has long been one of the most hostile places on the web, I’m not denying that. But this unprecedented drop in activity is 100% because of AI. Their product is as obsolete as horse-drawn carriages.
It was dying well before AI and while a knee happened coincidentally to chatgpt release, I don't think you remember correctly what gpt answer looked like for programmers questions back then. It was much later where it started assisting development.
A lot of people are rather bitter that they couldn't get the response they wanted from questions in the format they wanted to ask them, and would rather think that this slight against them is part of what killed the site. If it is all due to AI then they can't tell the tale of how SO rued the day its community deemed them off-topic :)
While AI was probably the last straw, it only hastened the collapse that was already underway.
LLMs so obviously make any type of the business model of SO unviable.
Attributing the death to mismanagement really does just seem like motivated reasoning due to hating SO.
StackOverflow was dead man walking as soon as some HR idiot at Google weaponized Fake Internet Points via quarterly reviews needing "social contributions".
Before that, the ignorant punters would leave these kinds of sites alone. Since there was nothing to be gained, you either had to be an altruist or deranged to waste your time on answering questions on the site.
After Google weaponized Fake Internet Points and everybody copied Google, a zillion software programmers everywhere in the world suddenly wanted the shortest path to earn lots of Fake Internet Points rather than actually answering relevant questions.
THAT was the death of StackOverflow--everything else has been the denouement.
--------
[0] I'm leaving tech as a career in the next year¹, I don't care if I have to wait tables on minimum wage, I've been unhappy since ~2020 for a growing number of reasons and the great purloining+cheating+polluting machines are the final bail of hay
[1] Maybe before then end of this year, my notice period is two months and I have some accrued annual leave to use, we'll see how the next week or so goes!
I'm not qualified for anything else that isn't also remote, full of AI and getting more so, or both, so I may quite literally end up waiting tables, I have friends who can help me skip the queue there with an “he's a good bloke” recommendation. While I was in hospital recently (gallbladder trouble and liver infection) talking to one of the porters who was wheeling to & from the CT scanner, they apparently need more bodies ATM, I could certainly do that (other than the temporary issue I was in there for I'm fit & healthy) and might actually find some enjoyment in it. Both involve dealing with real people who are in the room with me. Anything I do once I leave here will be result in a massive pay cut, but if the change saves my sanity then that will have to be acceptable.
I don't contest productivity benefits and there's a time and place for Claude in my workflow but I fear sooner or later they will start tracking if we fired up Claude for the day. Half joking. But it makes 1:1s even more awkward now when I have to put on a show for the GenAI parade.
Thanks. I might need it!
> sooner or later they will start tracking if we fired up Claude for the day
We track who uses what AI and how much. For billing/expense tracking purposes, as we've gone from “use it as much as you can” to “use it as much as you need to but don't waste it”, but there will be a table somewhere with my name in the bottom section next to a bit fat 0.
> has been pretty pushy when it comes to GenAI use tho … Flat out saying no comes with …
You'll have to deal with that or get out, eventually, how soon depends on your management. AI is not going away, even when/if the costs rise when all that investment capital starts being clawed back. If it is any help in choosing an exit strategy: under UK employment law (if you are elsewhere you'll have to verify how things work in your jurisdiction) they can't force you to use AI or punish/sack you for not as it can be argued to be a material change of role (as it is described as revolutionary, or in less breathy terms “like becoming a manager”, etc.) not just a different tool. But if you refuse then the same argument can be used to make you redundant (“we need more AI people and fewer traditional devs, traditional devs are being declared redundant, would you like to switch roles instead?” would be perfectly legal). As much as the payout for my long years in the same place would be nice, personally I don't think I can sit it out long enough for that to happen.
So yeah, AI surely didn't help, and I concede that the SO case is even worse given its nature (and how good of a substitute AI is), but if it wasn't a pool of acid I could see it existing as a place for people who want human help and opinions. So I don't think it's fair to say the drop is "100% because of AI", as you put it.
Then again, it's been a pool of acid since forever and it kept going for a long time like that, so maybe you are both right. AI did kill it, but because it was hostile.
Yeah, the place could be very hostile, but that hostility didn't crater questions before ChatGPT was released.
Also, they had the usual incentives to avoid doing things that would slow down initial growth. They were trying to build a community, and they definitely succeeded at that.
It took until the "middle times" for people to really develop an understanding of how questions for a Q&A site are supposed to look. And then the rest of the time was spent trying to convey those standards accurately, caught between the usual archetypes of angry forum users, in a place that was not supposed to be a forum.
I hear they're drowning in AI bots. I guess a fair amount of people also don't mind talking to those.
But SO was never intended to be a place where people "talk to each other". It was specifically intended and designed (sort of; the design was faulty) to avoid chatter because that gets in the way of others coming by later and finding a question and reading its answer.
AI isn’t magic and can’t always give you the right answer. And it can’t give you the why a lot of the time.
People’s actual experience is useful, and stack overflow was great for that. Until they ruined it.
If the site hadn’t been ruined, it still would’ve had value today. Traffic would’ve dropped, but not to near zero.
AI can’t always give you the right answer. But it tries. StackOverflow stopped trying to answer questions.
The product isn't obsolete. It's just been re-packaged by AI companies in a far more useful form.
Before AI people would still try asking questions on SO despite the hostility because it was the only/main place to get answers. Over this time, SO popularity was deteriorating.
After AI people who had bad experiences with asking questions on SO would ask the AI instead. This is because the AI wouldn't mark your question as a duplicate, wouldn't question its validity [1], etc. It would just answer your question. You could even ask it follow-up questions if the answer wasn't quite what you were looking for. All of this in real time, not waiting to get an answer.
[1] Unless the question triggered the safety gates that the AI companies have in place.
The same thing that happened to perl.
Perl likely declined because Perl6 did a Duke Nukem Forever, and Perl5 was dated and not incredibly well-adapted to OOP. Then Python got taken up for teaching, and it was all over.
(I still love Perl5, still use it in one-liners all the time, and still code up quick CLI tools with it.)
I used SO almost every day until capable LLMs showed up. I don't remember the last time I used it. SO and similar sites are dead, and LLMs killed them.
Damn sad how quickly that habit changed for me.
Never asked a question, but got some good info from people who had asked questions and the people who replied.