We Were Never Good Programmers

OK, that may be a little mean-spirited, but I don't just mean you. I mean a whole lot of us. A couple weeks ago, a graph made the rounds showing the decline of Stack Overflow. At its peak, there were 207,000 questions asked in a single month. By December 2025, there were just 3,600. That's a steep drop.

The accepted explanation is simple: LLMs have displaced Stack Overflow. And it's true, I rarely visit the site anymore. Many of my peers don't even know it exists.

A lot of people blame Stack Overflow's decline on heavy-handed moderation. There's plenty to criticize about how the site was moderated, but I think the story is more nuanced than that.

Stack Overflow never positioned itself as just another Q&A forum. It wasn't ExpertsExchange. The whole point was that it would be a place to find answers first, then ask questions if you couldn't find what you needed.

People were happy to point out when your question had already been answered. But this often frustrated newcomers. While waiting in line at the grocery store, during coffee breaks, anytime I had a spare moment on my phone while others were playing Candy Crush, I spent this time explaining to people that the problem wasn't that they couldn't find an answer. The problem was that they didn't understand their own problem in the first place.

When you see long comment threads where a poster is desperately trying to force-fit a solution into their problem, it's nearly impossible to convince them that the solution itself is wrong. The tools we had like closing questions, downvoting, often frustrated more than they helped. But moderation wasn't static. It evolved. It got better over time. The world of programming, however, changed faster.

Now most people, according to that graph at least, aren't asking their questions on Stack Overflow. They're going straight to ChatGPT or their favorite LLM. It may look like failure, but I think it means Stack Overflow did exactly what it set out to do. Maybe not the best business strategy for themselves, but the most logical outcome for users. Stack Overflow won by losing.

What users often struggled to understand is that their question wasn't unique. I'd argue that most programming questions have already been answered on Stack Overflow. Unless you're working with brand-new software and hitting an obscure edge case, your answer already exists somewhere.

When you ask ChatGPT that same question and it gives you an answer, it's not because the LLM is running your code and debugging it in real-time. It's because large language models are incredibly good at surfacing those existing answers. They were trained on Stack Overflow, after all.

When you ask an obscure C# question, ChatGPT isn't conjuring magic. It's just channeling Jon Skeet. In fact, I'm convinced all C# questions are secretly piped directly to his personal computer and he answers them in real-time...

Stack Overflow accomplished its mission of making programming answers accessible. A lot of people never experienced, or don't remember, what it was like to program before Stack Overflow. Getting answers on forums was a nightmare. Stack Overflow made the web a better place and created career opportunities for millions along the way. LLMs are eating its lunch now not because Stack Overflow failed to innovate, but because it succeeded in solving the problem.

But it's not all good news for programmers. For a lot of us, Stack Overflow forced reflection on our own skills. Knowing how to program is one thing. Knowing how to debug and solve problems is another.

Debugging requires knowing how to ask questions. On Stack Overflow, you had to ask a good question to get an answer. You needed to provide context, relevant background information, and demonstrate what you'd already tried. The discipline of formulating a proper question made you a better programmer. It forced you to understand your problem deeply enough to articulate it clearly.

Hence the title of this post. Maybe you know how to code, but do you know how to find answers to your problems?

When you use an LLM, it might give you an answer, but do you understand that answer? Do you know why it works? Can you tell when the LLM is wrong and might be introducing a new bug?

Those are the skills that atrophy when Stack Overflow fades away. Sure, you'll use AI agents to do your work. But when they can't solve the problem, and they will hit walls, can you take over? Do you have the skills to truly understand what's broken?

That's the question we should all be asking ourselves. With LLMs, you don't have to please a skeptical audience anymore. And you won't get the benefits of this friction.

---