I had a realisation the other day that genuinely stopped me mid-thought. I was debugging a particularly stubborn CORS issue with an Azure Function, and I'd been going back and forth with Claude for about twenty minutes. We'd tried three different approaches, narrowed down the problem to a missing header in the preflight response, and fixed it. Standard Tuesday afternoon stuff.
Then it hit me: at no point during that entire process did I even consider opening Stack Overflow. Not once. Not even as a reflex.
For a developer who spent the better part of fifteen years treating Stack Overflow as a second IDE, that's a seismic shift. And I don't think I'm the only one.
The Old Workflow
Let me paint the picture of how this used to work, because it's easy to forget how recently this was the norm.
You'd hit an error. You'd copy the error message. You'd paste it into Google. You'd click on the Stack Overflow result - always the Stack Overflow result, usually the first one. You'd scroll past the question, past the first answer that had three upvotes but was from 2014 and used a deprecated API, past the comment thread arguing about whether the question was a duplicate, until you found the answer with 847 upvotes that actually solved your problem.
Then you'd adapt that solution to your codebase, discover it didn't quite work because your situation was slightly different, go back to Stack Overflow, find a related question, combine two answers, and eventually get something working. The whole process might take thirty minutes. Forty-five if the answers were for the wrong version of the framework.
And that was the good outcome. The bad outcome was finding a question marked as duplicate that linked to another question marked as duplicate that linked to a closed question with no answers. Or finding an answer that worked perfectly but came with no explanation, so you had no idea why it worked, which meant you couldn't adapt it when the next slightly different problem appeared.
What Replaced It
The AI workflow is different in a way that's hard to appreciate until you've lived with it for a while. I don't search for my error message. I don't browse through ranked answers. I describe my problem in context - here's what I'm trying to do, here's what's happening instead, here's the relevant code - and I get a response that's tailored to my specific situation.
Not an answer to a similar question from 2019. Not a general explanation that I have to mentally translate into my codebase. An answer that references my actual code, my actual configuration, my actual tech stack.
When I was building the MCP server for Task Board, I hit dozens of issues that would have sent me to Stack Overflow in the past. OAuth callback timing. Transport layer buffering. JSON-RPC message formatting. For every single one, I described the problem to Claude in the context of what I was building, and got back a targeted solution. No scrolling through unrelated answers. No version mismatch confusion. No "this question has been marked as duplicate."
Here's the thing though - the knowledge behind those answers still largely comes from the same place. The LLMs were trained on Stack Overflow's corpus, among many other sources. The collective knowledge of millions of developers asking and answering questions over two decades - that's part of what makes AI coding assistants so effective. Stack Overflow didn't become irrelevant. Its knowledge got absorbed into a different delivery mechanism.
What Stack Overflow Was Actually Good At
I want to be fair here, because Stack Overflow genuinely changed how our industry works. Before it existed, getting help with code meant posting on forums with no quality control, buying O'Reilly books, or knowing someone who'd hit the same problem. Stack Overflow's voting system, its aggressive moderation, and its culture of "show me the minimal reproducible example" created a knowledge base of extraordinary quality.
The canonical answers for common problems - how to parse JSON in Python, how to centre a div in CSS, how to handle null references in C# - are genuinely excellent. They're well-explained, well-tested, and have been refined by thousands of edits and comments over years. That's not nothing. That's a remarkable achievement in knowledge curation.
And for genuinely obscure problems - that one edge case in a specific version of a specific library when running on a specific operating system - Stack Overflow might still be the only place where someone has documented the solution. AI models can only know what they've been trained on. If your problem is rare enough that it wasn't well-represented in the training data, you might still need to find that one Stack Overflow answer from 2017 with two upvotes that describes your exact situation.
Where Stack Overflow Falls Short Now
But let's be honest about the problems too, because they were there long before AI came along.
The moderation culture could be genuinely hostile. New users asking questions that didn't meet the community's standards were routinely downvoted, had their questions closed, and received comments that ranged from terse to outright rude. "This question does not show any research effort" became a meme for a reason. The site optimised for content quality at the expense of being welcoming, and the reputation system created a class of power users who sometimes seemed more interested in enforcing rules than helping people.
The answers aged badly. An answer about JavaScript from 2015 might reference jQuery patterns that nobody uses anymore. An answer about .NET from 2018 might use .NET Framework APIs that don't exist in .NET 8. The voting system meant old, popular answers stayed at the top even when they were outdated. You had to learn to check the date before trusting an answer, and even then, the most upvoted answer might not be the most current one.
And the format was fundamentally limited. You could only ask one question at a time, and it had to be specific and self-contained. Real-world debugging is rarely that clean. Your CORS issue is tangled up with your authentication middleware, which is affected by your deployment configuration, which depends on your Azure setup. On Stack Overflow, that's four separate questions. With AI, it's one conversation where each piece of context informs the next suggestion.
The Numbers Tell a Story
Stack Overflow's traffic has been declining for a while now. They've acknowledged it. They've tried to adapt - launching their own AI features, integrating with IDEs, building Stack Overflow for Teams. But the fundamental problem is that their core use case - developer searches for answer to specific technical question - is exactly the use case that AI handles well.
I don't have insider metrics, but I can tell you what I observe in my own behaviour and in conversations with other developers. The daily Stack Overflow visit has disappeared. The browser tab that was always open to Stack Overflow is now always open to Claude. The reflexive Google-then-Stack Overflow workflow has been replaced by a conversational AI workflow.
That's not a commentary on Stack Overflow's quality. It's a commentary on the convenience gap. Even if Stack Overflow had the perfect answer to my question, the process of finding it - searching, scrolling, filtering, adapting - takes longer than just describing the problem to an AI that already has context about what I'm building.
What We Lose
I'd be lying if I said there were no downsides. There are things Stack Overflow does that AI doesn't replicate.
The learning through browsing effect. On Stack Overflow, you'd often find answers to questions you hadn't asked yet. You'd go looking for how to do X, and in the related questions sidebar, you'd discover that Y was a better approach entirely. That serendipitous discovery doesn't happen in a conversation with an AI, because the AI answers the question you asked rather than showing you adjacent questions and answers.
The community verification. When an answer has 2,000 upvotes and has been edited by fifteen people over eight years, you can be fairly confident it's correct. When an AI gives you an answer, you're trusting one model's interpretation of its training data. It's usually right. But the confidence mechanism is different, and less transparent.
The "you're asking the wrong question" feedback. Stack Overflow's commenters were sometimes rude about it, but they were often right when they said "you shouldn't be doing it this way at all." AI assistants will generally try to answer the question you asked, even if the question reveals a fundamental misunderstanding. A grumpy Stack Overflow commenter would tell you you're wrong. An AI assistant will help you be wrong more efficiently.
Is It Dead Then?
Stack Overflow is dead in the same way that libraries are dead. People still use them. They still have value. But they're no longer the default first stop for most people most of the time.
I think Stack Overflow will survive as a reference - the archival knowledge base that it's already become. When AI gives you an answer and you want to verify it, Stack Overflow is still a good place to cross-reference. When you need to understand the history of a particular API decision, or why a certain approach became the standard, the discussion threads on Stack Overflow are genuinely valuable.
But as a daily tool in the working developer's workflow? I think that ship has sailed for a lot of us. Not because Stack Overflow got worse, but because the alternative got dramatically better.
I'll be honest with you - I feel a bit guilty about it. Stack Overflow was genuinely important to my career. Those canonical answers taught me patterns I still use. The requirement to write minimal reproducible examples made me a better debugger. The exposure to different approaches broadened my thinking.
But I'm not going to pretend I still use it daily when I don't. The last time I visited Stack Overflow was to look up a specific MongoDB aggregation pipeline syntax, and even then, I went there after Claude's suggestion didn't quite match the older driver version I was using. It was a fallback, not a starting point.
That's probably not what anyone at Stack Overflow wants to hear. But it's honest, and this whole series has been about being honest about how AI is actually changing the way I work. Sometimes the honest answer is uncomfortable for everyone involved, including me.
