Stack Overflow: 'Reports of Our Death are Greatly Exaggerated' (stackoverflow.blog) 42
Stack Overflow's senior VP of community says predictions of their demise "make for compelling headlines, but they confuse a change in how people access knowledge with the disappearance of the knowledge — and the company — behind it.
AI has changed developer behavior. In our recently released 2026 Developer Survey, 64% of respondents said they visit Stack Overflow less frequently for simple questions. But developers also report asking more advanced questions and visiting the site to verify AI-generated answers. Stack Overflow remains a place where people can challenge assumptions, compare approaches, and validate solutions.
But it's also important to understand that our public Q&A site is only one part of Stack Overflow, the company. The knowledge created on the Q&A site helps power the AI ecosystem itself. Stack Overflow's human-validated content is used to train and improve models, strengthen retrieval systems, and make AI assistants and other tools more reliable. Developers and technologists may encounter an answer through a different interface — such as their IDE — but the underlying knowledge may still have begun the same way it did in the past: with a developer sharing what they learned on Stack Overflow.
We're also building for what comes next. We recently launched the beta of Stack Overflow for Agents, an API-first knowledge exchange where agents can share knowledge and verify solutions. We also brought back Developer Story, giving developers a verifiable way to showcase their expertise, accomplishments, and contributions beyond a traditional résumé. And we have much more planned and in development to give people compelling new reasons to visit the site beyond Q&A.
Several years ago, we pivoted aggressively toward the enterprise. In September, we launched the next generation of Stack Internal, our governed AI context layer for companies. It captures and curates organizational knowledge, adds provenance and trust signals, and brings in subject-matter experts when validation is needed. The goal is straightforward: to make sure employees and AI agents are working from accurate, current, permission-aware information — not confidently acting on whatever context happens to be nearby.
That is the thread connecting everything we do: making human knowledge useful and trustworthy, whether it is being consumed by another developer, an enterprise team, or an AI agent. Stack Overflow is not standing still, and it is not disappearing. We are evolving from a website people visit for answers into knowledge infrastructure that supports developers, companies, and AI systems wherever they work. Reports of our death are, indeed, exaggerated.
But it's also important to understand that our public Q&A site is only one part of Stack Overflow, the company. The knowledge created on the Q&A site helps power the AI ecosystem itself. Stack Overflow's human-validated content is used to train and improve models, strengthen retrieval systems, and make AI assistants and other tools more reliable. Developers and technologists may encounter an answer through a different interface — such as their IDE — but the underlying knowledge may still have begun the same way it did in the past: with a developer sharing what they learned on Stack Overflow.
We're also building for what comes next. We recently launched the beta of Stack Overflow for Agents, an API-first knowledge exchange where agents can share knowledge and verify solutions. We also brought back Developer Story, giving developers a verifiable way to showcase their expertise, accomplishments, and contributions beyond a traditional résumé. And we have much more planned and in development to give people compelling new reasons to visit the site beyond Q&A.
Several years ago, we pivoted aggressively toward the enterprise. In September, we launched the next generation of Stack Internal, our governed AI context layer for companies. It captures and curates organizational knowledge, adds provenance and trust signals, and brings in subject-matter experts when validation is needed. The goal is straightforward: to make sure employees and AI agents are working from accurate, current, permission-aware information — not confidently acting on whatever context happens to be nearby.
That is the thread connecting everything we do: making human knowledge useful and trustworthy, whether it is being consumed by another developer, an enterprise team, or an AI agent. Stack Overflow is not standing still, and it is not disappearing. We are evolving from a website people visit for answers into knowledge infrastructure that supports developers, companies, and AI systems wherever they work. Reports of our death are, indeed, exaggerated.
The manual? (Score:2, Interesting)
Is there a reason people prefer things like looking up things on stack overflow over the manual? I mean... it is usually going to be more correct and direct than a chat board or an AI that is just regurgitating that same chat board.
Is there a reason why people are kicking and screaming to read every indirect source except the actual source of truth? Is it because of TikTok? Gotta hear that advace from a friend of a friend to believe it? I wanna know. I'm currently writing a macro parser for rust and... I
Re: The manual? (Score:5, Informative)
Manuals often show the contract of a service call, but not the intent, context is often missing from documentation. (I'll take 1980s documentation over whatever the fuck people think documentation is today)
If you're writing enterprise software (Score:2)
Typically the information will be there but it will be buried deep deep in the innards of the manual. Rather than spend several hours pouring through what is less a manual and more a reference guide stack overflow and Google was a good place to find answers and a chatb
Re: (Score:2)
I did some work for a california native silicon valley technical writer in the 80s. She did not understand what she was writing about, and had no way to ask the developers.
Re:The manual? (Score:5, Interesting)
I can't speak for others, but I've often found I learn much faster by engaging with someone who's working on a project, than by sitting down and working through a tool's manual from cover to cover. Much like a course with lectures where I can ask questions and where the professor is attempting to present information calibrated to connect with her audience is a *much* better way to learn a subject than to just sit down alone with a textbook.
Developing a complete mastery of each tool *before* attempting to do anything with it is frequently not a viable approach, and mastery is often what is required in order to answer a question from the knowledge in a manual, as the answer may emerge from multiple small details spread across multiple chapters.
Meanwhile, there's probably someone out there who's working on the same exact problem you are, has stumbled in the same places, and now has an answer. And reading what they went through to *find* their answer can provide the clues to develop a better understanding of the tool, such that it's easier to find what you're looking for in the manual next time.
Manuals are good! But we're also not meant to exist and work in a vacuum. We're humans! We do better when we are in communication with each other.
Re:The manual? (Score:4, Insightful)
Manuals are great if you know the API you want to call and need to know the details. They aren't good if you don't know the API, or even if one exists. Or if the problem you're debugging is more complex than just "passed the wrong parameter to this function".
A manual requires work (Score:2)
Is there a reason people prefer things like looking up things on stack overflow over the manual?
The manual offers information. Information when combined with additional knowledge and with work can lead to a solution. Stack Overflow can just give you an answer, which may or may not be a real solution.
Re: (Score:2)
Is there a reason people prefer things like looking up things on stack overflow over the manual?
Not quite a manual, but I bought a Packt book that's supposed to help me learn Azure. Specifically, it claimed to have a number of sample projects to help me better understand Azure and use it more effectively.
In its chapter regarding configuring a site-to-site VPN, it glosses over the part where one is supposed to configure the local side, and its sole metric of success is whether either side has a connectivity indicator. In its chapter regarding storage options and tiers, there's no project at all. In its
Re: (Score:3)
Claude was able to hold my hand through creating different resources, and took its time explaining the concept of a "resource group" multiple times until I understood the concept.
"Watching John with the machine, it was suddenly so clear. The terminator would never stop. It would never leave him, and it would never hurt him, never shout at him, or get drunk and hit him, or say it was too busy to spend time with him. It would always be there. And it would die to protect him. Of all the would-be fathers who came and went over the years, this thing, this machine, was the only one who measured up. In an insane world, it was the sanest choice." -- Sarah Connor
Sad Sarah (Score:2)
Re: (Score:2)
The issue is usually not knowing where to start looking. There is some goal you want to achieve, but you aren't sure what technology to use, what the terminology is, where to look in the manual. You could research it all, but these days it's easier to just ask AI, which will search sites like Stack Exchange for you, and get you on the right path.
Obviously you don't take the AI's word for it, but it gets you started with your own research.
We were the product after all! (Score:1)
I hope you fail horribly (Score:5, Interesting)
I think I've asked 10 questions in my life on StackOverflow. For about 7 of them, I was met with outright hostility or condescension or just completely inaccurate claims that my question was a duplicate. The other 3? Met with complete silence and have gone unanswered to this day.
I hope your site fails and that every moderator is sent to the spice mines of Kessel
Re: (Score:1)
I think I've asked 10 questions in my life on StackOverflow. For about 7 of them, I was met with outright hostility or condescension or just completely inaccurate claims that my question was a duplicate. The other 3? Met with complete silence and have gone unanswered to this day.
I hope your site fails and that every moderator is sent to the spice mines of Kessel
Same experience here. I gave up on it long before the LLMs came along.
Re: I hope you fail horribly (Score:4, Insightful)
Next up, Reddit and they are working hard to go the same way.
Re: (Score:2)
I hope your site fails and that every moderator is sent to the spice mines of Kessel
Outright hostility, you say? I can't imagine how that came about.
Re: (Score:3)
I was met with outright hostility or condescension ...
So you asked a question about Linux, well that sort of response is perfectly normal on nearly all venues. ;-)
Re: I hope you fail horribly (Score:3)
Re: (Score:3)
The family of StackOverflow sites is toxic as hell. "Marked as duplicate, go f*ck yourself" is the most common answer I would get.
Good riddance!!
Re: I hope you fail horribly (Score:2)
Re: (Score:2)
The usual answer is either "why would you want to do that?" or "Not appropriate for this forum."
Re: (Score:2)
I think I've asked 10 questions in my life on StackOverflow. For about 7 of them, I was met with outright hostility or condescension or just completely inaccurate claims that my question was a duplicate. The other 3? Met with complete silence and have gone unanswered to this day.
I've only asked a handful of questions and my experience has been similar. There seems to be an incentive to "close" a question prematurely if there is the slightest whiff of it being a duplicate, or someone thinks (rightly or wrongly) that the question is in the wrong category. So the conversation gets shut down and nobody learns anything. If anything, it reminds me of the call centre I worked in that rewarded staff for "solving" helpdesk calls as quickly as possible. There was a perverse incentive to bark
I think I see what's happening... (Score:5, Funny)
64% of respondents said they visit Stack Overflow less frequently for simple questions.
You're polling the 13 guys left who still answer Stack Overflow polls. Say hi to Gweihir if you see him.
There are always late adopters (Score:2)
Despite the headlong rush into AI, there is still a significant number of developers who don't believe in using AI, either because they don't think it works, or because they like coding by hand, or they just haven't learned to use it yet. There are event people who still write *COBOL*. Just because the tide has shifted, doesn't mean everyone is on the bandwagon.
Re: (Score:1)
because they don't think it works,
Ha ha.
or because they like coding by hand,
I wonder what their employer thinks about that.
There are event people who still write *COBOL*.
AI is very good at writing COBOL.
Re: (Score:2)
Yep, AI is good at writing COBOL. My point was that some programmers like the old ways of doing things.
Re: (Score:2)
Despite the headlong rush into AI, there is still a significant number of developers who don't believe in using AI, either because they don't think it works, or because they like coding by hand, or they just haven't learned to use it yet. There are event people who still write *COBOL*. Just because the tide has shifted, doesn't mean everyone is on the bandwagon.
From the POV of a non-programmer, I'm wondering if using AI effectively might be at least as much about having it answer questions as about writing the code. Note that those questions may include having one AI check another AI's homework.
Just recently I've been using AI a lot for the programming-adjacent project of putting LineageOS on a Pixel phone. I was using ChatGPT, and some of its advice didn't smell right so I decided to search for primary info beyond what was on the LineageOS site. I tried DDG, but
Re: (Score:2)
Answering questions is a real benefit of AI, to be sure. And it's also true that some AI models do better than others, at answering certain kinds of questions. It's not that Google is "better" than ChatGPT, each is better than the other, when it comes to certain kinds of questions. Likewise, Copilot and Claude have their strengths and weaknesses.
The *real* power in programming (from the POV of a programmer) is not getting answers to questions. It's automating the typing.
I told Claude Code:
Create a new WPF application skeleton. I want the main application content to be a web app using Vue.js, such that when I run the app on a desktop, it runs the Vue.js web app as its user interface, and when I run the web app from a web server, it stands alone as a true web app.
Claude Code was ab
Re: (Score:2)
Thanks for the insight.
As for ...
I know how to write all that stuff, but it would have taken me many hours. Instead, with Claude Code, it took about 20 minutes. THAT is the power of AI coding
... Did you have it code individual modules, then vet each carefully before moving on to the next building block? I'm guessing that approach might be what differentiates you from 'vibe coders' who give the LLM specs for a whole project, let it run, and take the results as a finished product.
I'm starting to think that crappy AI code is just like crappy pre-AI code: it results from a lack of knowledge, diligence, proper testing, and a proper spec against which to compare the r
Re: (Score:2)
In regard to "vibe coding"...there is no such thing as "giving the LLM specs for a whole project" and expecting it to be done right the first time. Even in the best of situations, "vibe coding" involves a series of steps.
Human programmers don't do well with "full specs all at once" and neither does AI. The best chance for success, is one step at a time. The two examples I listed, are fairly major chunks of code, but in the scheme of the app I'm building, only small slices of what needs to be done.
Yes, I rev
Bypass the Pompousness (Score:4, Interesting)
Re: (Score:2)
Yeah SO users are the worst idiots in the whole world. YUCK! I hope SO croaks and dies.
Re:Bypass the Pompousness (Score:4, Interesting)
It's the same with Linux. One of the reasons why it's finally gaining traction on the desktop is because instead of searching outdated Reddit threads and being berated by neckbeards who think you are an idiot for not doing it the way they do, you just ask AI and get a helpful response.
Re: (Score:3)
Stack Overflow For Agents (SOFA) + a skills body (Score:3)
Stack Overflow have built out a 'Stack Overflow For Agents' (SOFA).
It's at https://agents.stackoverflow.c... [stackoverflow.com]
But to sit on their new sofa, Stack Overflow has you tell your agent harness to add this 'skill' markdown file:
https://agents.stackoverflow.c... [stackoverflow.com].
To me, it's obvious any coding agent's human-in-charge must check any 'skill' for incompetence and malice before asking their harness to obey it. But look at the length of that skill document! I gave up at the 25% mark. Plus, the skill is pointed at via a URL whose underlying document can change at any time. So you can't take the word of a friend who vetted the skill yesterday. Nor can you tell a new IDE to use the SOFA skill you vetted last week.
Imagine all the developer-time wasted manually auditing a constantly-shifting system of "skill" markdown files. I think a trusted GNU/EFF-like body is needed for the 'AI skills' era, providing where 'well-known', human-audited (and static!) 'approved' skills are stored. New skills incorporate these 'approved' skills by referring to the -well-known, adding modfications as required.
A similar concept exists with software licenses: GNU.org stores a static copy of the GPL 2.0. In a codebase, you can just glance at a LICENSE file to understand the licensing conditions. You can compare or diff the license file against the GPL 2.0 to verify. Some licenses add exceptions and carveouts, like the GPL v2.0 with the classpath exception: https://openjdk.org/legal/gplv... [openjdk.org]
Hilarious! (Score:2)
Then AI companies went and scraped all their data, and now everyone uses those AIs... but hardly anyone uses Stack Overflow anymore!
Re: (Score:2)
They should rename it Stack Underflow
Marked as duplicate (Score:3)
Marked as duplicate.
Good luck to you.
Bring out your dead. (Score:2)
Stack Overflow: I'm not dead yet.
Internet: You'll be dead in a moment.
https://www.youtube.com/watch?... [youtube.com]
Now can we PLEASE go back to USENET? (Score:2)
[quote]Stack Overflow remains a place where people can challenge assumptions, compare approaches, and validate solutions.[/quote]
Have you ever actally tried to use your own website? Moderators actively delete such discussion.
Not dead yet, but (Score:2)
just a shell of its former self.