Software application developer code advice portal Stack Overflow has reportedly experienced a significant downturn in usage. The December 2025 “questions posed” figures saw 3,862 queries raised, a 78 percent drop compared to the same month a year before. With AI services now fulfilling developers’ query needs, now is the time to ask… is Stack Overflow drying up and what lessons should we take from this change?
It’s easy to point fingers here and suggest that software engineers are naturally turning to AI services to answer questions. Given that they can now do so from within their chosen Integrated Development Environment (IDE), it makes logical sense to assume this might happen. But there are other factors in the mix here; over-critical site policing has also been highlighted.
Having been called out for its dismissive and overly negative moderators who deliver punitive remarks on what might be perceived as “stupid questions” by some on Stack Overflow, the universe of AI chatbots and code-bots is clearly less negative.
Punative policing policy
As Peter Huber writes on Reddit, “Stack Overflow lives from the contribution of thousands of contributors, but is controlled by a small band of moderators who have dictatorship-like powers. They probably spend a significant part of their life on Stack Overflow, which allows them to collect enough points to get the power to do as they please. Of course, it is good that Stack Overflow tries to enforce a certain level of quality for its content, but having this ‘police’ [tactic] punishing others leads to the same problem as any other police [body]: they feel they are omnipotent and always right, even if they are wrong.”
Huber further states that he wishes Stack Overflow would be more supportive of its content providers and help them to improve their contributions instead of removing content with wrong justifications, leaving no easy way to challenge decisions.
Duplicitous duplicates?
He further suggests that one “tool” (by which he means technique) that moderators like to use to suppress a discussion is to mark a question as a duplicate, which means the question/answers get closed and no further discussion is possible.
Stack Overflow, meanwhile, is still upbeat; the site tells us that it “strives to be the most vital source” for technologists, helping them to cultivate community, power learning, and unlock growth.
“Millions of the world’s developers and technologists visit Stack Overflow’s public platform to ask questions, learn and share technical knowledge, making it one of the most integral websites in the world with over 83 million questions asked and answered. Stack Overflow’s enterprise knowledge ecosystem, Stack Internal, is the go-to space that 20,000 organisations turn to for validated expertise so that teams can accelerate productivity, reduce enterprise risk, and leverage AI with confidence,” notes the company’s mission statement pages.
Unification situation
The issue may of course be more directly related to the fact that major cloud hyperscalers, weighty enterprise software/data players and every other vendor worth their salt in the new neo-cloud economy all want to keep developers on their single unified platform offering. It’s the whole promise of one workflow, one toolset, with one unified infrastructure beneath you.
Stack Overflow is free to use for bona fide software application developers and also operates a paid-for version of enterprise use. This distinction means that a drop in the number of questions raised does not necessarily equate to a drop in revenue and the site’s owners have reported revenue growth last year.
“Our edtech businesses demonstrated resilience and adaptability through disciplined execution. Stack Overflow and GoodHabitz achieved revenue growth of 12% (9%) to US$95m, primarily driven by Stack Overflow’s performance,” notes Stack Overflow owner Prosus, in its 2025 financial performance report.
A perfect stack storm?
Taking all this together, has the combination of factors here, including the onus on having to “present your question” openly on an open forum, mixed with the inevitable ease-of-use offered by AI now created a perfect storm?
Mandi Walls, DevOps advocate at AI-powered digital operations platform company PagerDuty has opinions in this space and she says that Stack Overflow certainly “looks like it is drying up” these days.
“AI developer tools seem to be taking attention away from static question-and-answer solutions, replacing Stack Overflow with generated code without the middleman… and without waiting for a question to be answered,” said Walls. “Interestingly, AI tools lack the reputational metadata that Stack Overflow relied on: i.e. when was this solution posted and who posted it… and do they have a lot of prior answers? Developers are conferring trust to LLMs that human-sourced sites had to build over years and fight to retain. It’s much easier for developers to ask an agent for some code to accomplish a task and click accept, regardless of the provenance of that code.”
Peter Zaitsev, founder at Percona says he’s not surprised either. But, he advises, we (i.e the software engineering community at large) do still need a place for conversations around real innovation, often so that we can discuss material that has not been created and used in training collateral.
“Today we know that LLMs like ChatGPT are already pretty good at answering common questions, which are the bulk of the questions asked at StackOverflow. Additionally, LLMs can respond in real time, so it is not a surprise that people were shifting away from StackOverflow. It might be not the only reason though – some people also reported StackOverflow moderators being rather hostile and unwelcoming towards new users, which had additional impact,” said Zaitsev. “Why would you deal with what you see as bad treatment, if an alternative exists?”
On the soul-less robot train
He points out that it’s interesting to look at how StackOverflow compares to Reddit, which has reported significant traffic growth, whether this is despite or because of the rise of LLMs. He thinks the difference may come from Reddit’s focus on specific discussions topics that engineers are passionate about having with other humans, rather than getting answers to technical questions where a “soul-less robot” response is just as fine.
“I mentioned LLMs are good at answering common questions as they are trained and can pull from existing content. They are however, not great at answering problems which were not solved before,” advised Zaitsev. “All you can get in these circumstances is speculation and hallucinations. We still need a place for those questions that require original research and investigations to be answered. If those kinds of conversations don’t take place on StackOverflow, then they will need to find another place to exist as well.”
Kennith Jackson is senior vice president of solutions at Andela, one of the world’s largest marketplaces for technical talent. He thinks that Stack Overflow’s recent dip in public questions reflects the evolution of developer learning habits.
“With AI now available directly in IDEs, engineers naturally turn to quick, contextual support as they work,” said Jackson. “Even so, communities like Stack Overflow continue to play an important role by offering the depth, perspective and shared experience that help developers understand not just what to do, but why it works.”
A blend speed… with depth
He added, “As AI and human expertise increasingly complement each other, the platforms that remain central to learning will be those that blend speed with depth and make it easy for developers to build confidence and contribute. We see this shift as a positive step toward a more holistic learning ecosystem, one where AI accelerates exploration and communities like Stack Overflow reinforce accuracy, mastery, and long-term growth.”
Jackson’s words may be the most balanced and insightful (and somewhat more optimistic) so far. Rather more fatalistic is Soham Mazumdar, CEO and co-founder of WisdomAI, a company that turns complex data into natural language insights and generates reports.
“Stack Overflow will probably die out. LLM-powered IDEs can already answer the vast majority of questions, and because they’re directly connected to a developer’s codebase, they have far more context than anyone could realistically share on Stack Overflow. That makes them a more natural place for developers to look first when they’re trying to solve problems in real time,” said Mazumdar.
Carter Page, EVP of R&D at Astronomer is of the same mind.
“LLMs, whether IDE-enabled or not, are clearly reducing reliance on Stack Overflow. Earlier foundation models depended heavily on training data from Stack Overflow – I’ve even seen an LLM reproduce one of my own answers verbatim. As these models are increasingly paired with more sophisticated logic and reasoning layers, they’re able to derive novel solutions rather than simply regurgitate past ones. Taken together, it’s hard not to see Stack Overflow as being in its last days” advised Page.
Did a human write this?
All of which leads us to (hopefully) a better understanding of what’s happening to a much-loved software application development resource in the age of artificial intelligence. All that’s left (perhaps) is the need to point out that this compendium of industry luminary opinions was not collated by a machine; it actually required real human interactions by a journalist, who also filtered and editorialised the end result to present this analysis as shown.
So at least that element of human work, with empathy and judgement, will never change, right?