While AI has created many massive new companies, it has also brutally all but ended some others.

StackOverflow is perhaps the starkest example. For most of the 2010s, it was the backbone of how software got built — not just a website, but genuine infrastructure for the global developer community. Launched in 2008 by Jeff Atwood and Joel Spolsky, the platform ran on a brilliantly simple premise: developers posted technical questions, peers answered them, and the community voted the best solutions to the top. Over time, it accumulated tens of millions of answers across every conceivable programming problem, becoming the de facto external memory of the software industry. If you were stuck, you went to StackOverflow. That was just how it worked.

The chart of questions asked on the platform tells the whole story in a single curve. From 2008 onwards, question volume climbed sharply — hitting around 100,000 per month by 2011, then pushing past 200,000 by 2014. Through the mid-to-late 2010s, it held relatively steady in that band, oscillating between 150,000 and 200,000 monthly questions. There were dips and spikes, but the platform remained one of the most visited technical resources on the internet. Then, right around late 2022, the line simply falls off a cliff.

By late 2025, monthly question volume had dropped to around 20,000 — roughly a 90% collapse from peak. The timing is not coincidental. ChatGPT launched in November 2022. GitHub Copilot became genuinely capable around the same period. Claude, Cursor, and a growing ecosystem of AI coding tools followed. Developers who had spent years reflexively opening a new browser tab and typing their error message into Google — and landing on StackOverflow — now had something faster, more personal, and far less judgmental to turn to.

The practical calculus changed completely. Posting a question on StackOverflow meant framing the problem clearly enough for strangers to understand it, waiting for responses that might or might not apply to your specific situation, and sifting through multiple answers of varying quality. An AI coding assistant does none of that. You describe the problem in plain language, get working code within seconds, and iterate from there. The AI won’t close your question as a duplicate. It won’t tell you to read the documentation. It will just answer.

A study examining StackOverflow’s post-ChatGPT decline found that the drop wasn’t limited to simple or low-effort questions — ChatGPT appeared to be displacing high-quality contributions as well. The impact was also sharpest for the most popular programming languages like Python and JavaScript, which makes sense: the more code exists publicly for a language, the better AI models are trained on it, and the more reliably they can answer questions about it. StackOverflow’s most trafficked territory became precisely the territory where AI was strongest.

The company’s business reflected this. In October 2023, StackOverflow laid off 28% of its workforce, its second round of cuts that year, after traffic to the site had already fallen noticeably following ChatGPT’s release. CEO Prashanth Chandrasekar acknowledged the need to “spend less” without directly naming AI as the cause — though the timing made the connection hard to ignore.

There is a particularly sharp irony embedded in all of this. StackOverflow’s enormous repository of questions and answers — accumulated over 15 years of developer contributions — was part of the training data that helped make large language models so capable at answering programming questions. The platform essentially helped train the tools that made the platform unnecessary. Its own community’s collective knowledge was used to build a replacement.

The broader shift in how developers work has accelerated this further. Vibe coding — the practice of describing what you want built in natural language and letting AI generate the code — has moved from novelty to mainstream workflow. Experienced developers are using AI to move faster; people who wouldn’t previously have called themselves programmers are building functional applications. In this environment, the friction involved in posting to a public forum feels like a relic from a different era.

None of this means StackOverflow’s archive has no value — it remains a vast, indexed record of how programmers solved problems across a specific period of technological history. But as a living, active community where developers go to get answers in real time, it has been largely displaced. The chart makes that plain. A platform that once handled hundreds of thousands of questions a month is now handling a fraction of that, and the trajectory shows no signs of reversing.

AI has a way of making the previous generation of tools look not just outdated, but almost quaint. StackOverflow was enormously valuable for more than a decade. The speed with which that value evaporated is a reminder of how completely AI is reshaping the tools developers rely on — and how little time it takes for something indispensable to become optional.