Stack Overflow's Decline

Stack Overflow recently posted information about how many new questions users ask. On the graph, which shows all the numbers from the very beginning of the platform until now, we reached in December 2024 a historic point: the total number of new questions was quite similar to the one in the first ever month of Stack Overflow. That's a pretty spectacular decline. In fact, during the Corona pandemic there was a steep rise in that category, but with the introduction of LLMs the trend has reversed, and it's still going downward.

Meanwhile, Stack Overflow laid off 28% of its staff to cut losses, and it seems to be changing the direction in which the company wants to go. As its CEO said recently, Stack Overflow's focus should move from easy questions, which are ideal to be answered by LLMs, toward more complicated and complex problems where users' know-how and experience matter the most. Besides that, Stack Overflow made deals with top AI companies to train models on its human-generated content, similar to how Reddit handled it.

Platform Issues: Duplicates and Toxicity

For me personally, it's neither bad nor good news. I was somewhat active on the platform around 2017-2019, at the beginning of my career as a web developer. Back then, Stack Overflow was definitely the main source of this kind of very specific and narrow knowledge about edge cases and subtleties in real production-ready code, where official docs were completely inefficient or non-existent, like in the case of CSS.

But since that time, the situation has changed drastically. In my experience, as ECMAScript started evolving and browser engines started reducing differences between themselves, the entire need for Stack Overflow was in question, as many edge cases were not there anymore due to long-awaited browser compatibility.

Additionally, the platform started to have another issue: closing new questions as soon as they got marked as duplicates, which in many cases was rather controversial, and a steadily increasing level of toxicity between users, so that many of them were completely discouraged from asking new questions. To be fair, I never experienced this on my own to such a degree that I could say it's something abnormal and visible only on Stack Overflow.

After a while, we were stuck with aging answers for more and more obsolete problems. With all of that, no wonder that Stack Overflow, at least for me, became a place I didn't want to visit—at least not actively and not necessarily as a contributor. My focus shifted entirely toward finding answers for my current issues I'd just encountered, which is kind of sad.

Can Stack Overflow's Pivot Actually Work?

As of now, it seems that LLMs are the way to go for most cases. They're faster and in most cases definitely better suited for one's specific use case. On the negative side, if someone did solve something non-trivial, it gets lost in one's chat. No one else has access to it, and perhaps the one owning the chat will, after some time, ask the same question once more instead of searching in the chat history, as it would be way easier and quicker than trying to find that one small thing from all the noise.

I'm somewhat interested whether that strategic shift Stack Overflow's CEO has mentioned will really take place. It could be beneficial for users to have some fixed database where one could find answers to more deep and hard questions, but there are two potential issues I can see right away:

1. Someone needs to ask those kinds of questions, and
2. Someone needs to be able and, for the most part, willing to answer them.

Currently, I would rather say that even those kinds of problems would be better handled by LLMs or by asking on Reddit or even Discord. I just don't believe in such a spectacular shift in what Stack Overflow is and how its users tend to use it.