Stack Overflow: When We Stop Asking | CSS-Tricks

Let’s play a quick game: I’ll show a graph and try to guess what it’s about.

Source: Data Stack Exchange

No, it isn’t a crypto coin crashing a few hours after being minted. And not, it is also not an oscillatory/wavy graph made with pure CSS, but a harsher truth.

I already gave it away with the title, but it still hits like a ton of bricks to know it is the steep decline in the number of questions asked on Stack Overflow . You can see its peak around 2014 with more than 200,000 questions asked in a single month. But now in 2026, it is struggling to even hit 3,000 questions a month.

We don’t have to be experts in the field to find out the culprit. You guessed, it’s AI… mostly.

While AI is painted as the Stack Overflow killer, the truth is Stack Overflow’s downfall started long before ChatGPT’s release in late 2022.

By community accounts and also from personal experience, moderation since its peak in 2014 has been (and still is) one of the leading causes for the lack of questions.

As the site grew, Stack Overflow needed a better way to moderate the hundreds of thousands of questions asked every month: the inevitable wall that forum-based communities hit when they scale beyond a certain point. There are several ways to try to solve this, but the route Stack Overflow took might not have been the best:

On Stack Overflow, we close or delete questions that can’t be answered straight away – it’s not very sociable, but it scales wonderfully.

It’s clear Stack Overflow wasn’t focusing on the quantity of the questions but rather on the quality of them, while avoiding duplicates as much as possible. This pattern was in favor of Google searches for questions that were already answered and, hence, living on pre-answered questions instead of on users making new or duplicate ones.

It wasn’t helpful either how the community seemed to close upon itself, making it harder for beginners to even ask a question. And if you’re like me, you probably want to inquire without being told you’re stupid, as if getting punished for wanting to learn.

Generative AI was the final nail in the coffin. I can’t complain about this, as AI seemingly provides the same answers without judgment (in fact, maybe too much encouragement) nor delay, so I can see why people might prefer asking an LLM instead.

However, as I dug deeper into this, my concern was no longer about just Stack Overflow, but the tech ecosystem at large. Questions like, are we still asking questions? Are we still seeking to be better? Or do we all rely on LLMs, and solely on LLMs, for advice? That kept ringing in my mind as I continued my research.

I believe that, beyond the fall of Stack Overflow, those questions linger more than ever. How AI has generally impacted our workflow, how we can use it in problem-solving, and what we can do about this as developers.

Problem-Solving and AI

Is AI a better programmer than you? What makes a programmer better than others is as subjective as it gets, but some are eager to say that AI can write code better than you. According to that research:

AlphaCode achieves human-level problem solving skills and code writing ability as shown by performance in programming competitions.

At least that’s when it was tested against Codeforce’s (an online code competition site) problems, where I admit it can and will perform better than your average programmer. But most developers don’t care about Contest problems beyond a technical interview; they know being a software developer is so much more than that.

AI writing quality code is an extremely nuanced topic and lacks a decisive conclusion. However, if you take the time to research, you’ll find that AI-generated code has lots of flagrant differences. According to the research from Cornell:

AI-generated code is generally simpler and more repetitive, yet more prone to unused constructs and hardcoded debugging, while human-written code exhibits greater structural complexity and a higher concentration of maintainability issues.

Okay, so it can generate simple code, but can it write good code? Even solve problems better than a software engineer would?

According to MIT research, AI can write good code, but it cannot possibly think and make decisions like a software engineer. AI cannot compete on that level yet, at least without running into a lot of bugs.

Drawing on both first-hand experience and feedback, if all you do is copy-and-paste AI-generated code without careful consideration, you are bound to hit serious bugs and possibly even vulnerabilities. In fact, VeraCode published an article stating that "[…] 45% of AI-generated code contains security flaws," after testing for security vulnerabilities in 100 AI models. That’s a large percentage of code that’s flawed security-wise and would have cost implications for any user who wants to "vibe-code" without doing thorough checks.

Fun fact: GitHub released the results of its AI in software...