The site the whole industry copies and pastes from

There is a small, quiet ritual that runs underneath most of the software you touch every day. A developer hits a wall - an error message they've never seen, a function that refuses to behave - and they open a browser tab. Nine times out of ten, the answer is already sitting on Stack Overflow, written years ago by a stranger, argued over by a dozen more, and stamped with a green checkmark by someone who decided it was correct. That checkmark is the product. Everything else is scaffolding around it.

Stack Overflow, operated by Stack Exchange, Inc., is the public question-and-answer platform where more than 100 million people a month go to solve coding problems. It holds roughly 83 million questions and answers, a library the company says has been leaned on some 113 billion times. Owned by the technology investor Prosus since a $1.8 billion acquisition in 2021, it is one of the most-visited sites on the internet - and, in the middle of an AI boom that was supposed to make it obsolete, one of the more interesting businesses in software.

01 / ORIGINSTwo programmers and a broken internet

In 2008, Jeff Atwood and Joel Spolsky - both well-known developers who wrote widely-read blogs - started recording weekly calls about a product they wanted to build. The premise was blunt: the internet, as it existed for programmers, was bad. Coding answers were locked behind paywalls, buried in expired forum threads, or scattered across sites optimized for ads rather than accuracy. Their fix was a Q&A site with two unusual ideas welded to the front of it. Answers could be voted up or down by anyone. And the best answer could be marked as accepted, so you didn't have to read the whole thread to know which reply actually worked.

Those two mechanics did something subtle. They turned a message board into a ranking system - a way of sorting not just what people said, but which version of the truth the community had settled on. The founders never planned to keep the recordings; the podcast that documented the build survives as an accidental oral history of how the thing was made. Early team members Geoff Dalgas and Jarrod Dixon joined soon after, and by 2010 the two-person experiment had generalized into the broader Stack Exchange network - today more than 170 community sites covering everything from mathematics to home repair, all running the same voting-and-reputation engine.

The company grew through the usual venture path - early rounds from Union Square Ventures and Index Ventures, later capital from Andreessen Horowitz - but the through-line was never the funding. It was the accumulation. Every answered question made the next search more likely to succeed, which pulled in more developers, which produced more answers. That compounding is boring to describe and very hard to replicate, which is roughly the point.

02 / THE MACHINEReputation as currency

Here is the part worth stealing if you are building any community. Stack Overflow never paid the people who made it valuable. It paid them in reputation - a public score that goes up when your answers get upvoted and unlocks privileges as it climbs. Status, it turns out, is a currency you can print for free, and millions of developers have spent nights answering strangers' questions to earn it. A new question lands on the site roughly every 21 seconds, and the whole apparatus of voting, editing, and sometimes-brutal moderation exists to keep the signal clean.

That design is also the moat. Roughly 82% of developers visit the site multiple times a month, according to the company. Habit is hard to compete with, and a decade and a half of accumulated, cross-linked, human-checked answers is nearly impossible to rebuild from scratch. It is the difference between a library and a pile of books.

03 / THE BUSINESSHow a free site makes money

The public platform is free, which raises the obvious question of how any of this pays for itself. The answer is a set of products built around the audience and the archive. The biggest is Stack Internal - until early 2025 called Stack Overflow for Teams - a private, paid version of the Q&A engine that companies run behind their own walls. Instead of losing knowledge every time an engineer leaves or goes on vacation, an organization captures it the same way the public site does: someone asks, someone answers, and the answer stays searchable. More than 15,000 companies use it, among them Microsoft, Instacart, Box, Chevron, and Siemens.

Around those sit an advertising business - Stack Ads, aimed at brands trying to reach developers, an audience notoriously allergic to being marketed to - and a newer, more consequential line: data licensing. Stack Overflow now sells structured access to its knowledge base to the companies training large language models. It is a genuinely strange arrangement, and the most telling one.

04 / THE AI QUESTIONSelling the pickaxe to the gold rush

When ChatGPT and coding assistants arrived, the common prediction was that they would kill Stack Overflow. Why scroll through a thread when a chatbot answers instantly? Traffic pressure is real. But the company's response was less about defense than about repositioning. In 2023 it introduced OverflowAI, which surfaces AI-assisted answers drawn from - and attributed to - the existing knowledge base rather than generated from nowhere. Then it started licensing that base to the model builders themselves, with partnerships that have included OpenAI and Google Cloud.

The logic sits in the company's own data. Its 2025 Developer Survey, drawn from roughly 49,000 responses, found that more than 80% of developers now use AI tools - and that trust in AI-generated answers had fallen to about 30%, down from roughly 40% a year earlier. Two-thirds said they were frustrated by solutions that are "almost right, but not quite," the kind that cost more time to debug than to write.

That gap - fast machines, distrusted output - is the whole pitch. If AI writes the code but a human still has to decide whether it's right, then a corpus of human-verified knowledge becomes the thing you check against. Stack Overflow's bet is that it can be the trust layer sitting inside the AI stack: the source the models quote, and the place people go when the model is confidently wrong. Prosus, which folded the company into its global education-technology portfolio, paid $1.8 billion for essentially that - 18 years of arguments developers had so that no one has to have them again.

05 / THE MARKETWhere it fits, and who's circling

Stack Overflow occupies an unusual position: part reference work, part social network, part enterprise software vendor, and now part data supplier. Its competition is diffuse. GitHub is nearby with Discussions and Copilot. Reddit, Discord, and Slack communities absorb some of the conversation that once went to Q&A threads. Inside companies, Confluence, Notion, and Guru fight for the same internal-knowledge budget that Stack Internal wants. And the AI assistants - Copilot, ChatGPT, Claude - are both a threat to the front door and a customer at the back one.

What none of them have is the specific thing Stack Overflow spent 18 years building: a ranked, attributed, community-vetted record of how software actually breaks and gets fixed, updated every 21 seconds. That archive is difficult to buy and slow to grow. It is why a site that some keep writing obituaries for keeps showing up in the training data of the tools meant to replace it.

06 / THE PEOPLEWho runs it now

The founders have long since stepped back. Prashanth Chandrasekar took over as chief executive in 2019, arriving from Rackspace, and it was on his watch that the company raised its $85 million Series E and then sold to Prosus. His public message has stayed consistent through the AI shift: knowledge is only useful if you can trust where it came from, and speed without attribution is a liability, not a feature. It is a convenient argument for a company sitting on the internet's largest pile of attributed answers - but the survey data suggests it is also just true.

The expertise inside the company is unusual in that a good deal of it lives outside the company. The moderators, the high-reputation answerers, the people who edit a badly-worded question into something searchable - most of them are volunteers, and the culture of the business is built to serve and occasionally to referee them. That relationship has not always been smooth; developers who love the archive will happily complain about the moderation that keeps it clean. But the friction is the mechanism. A place where anyone can post and nobody edits would not be worth 113 billion lookups.