In January 2026, a developer named Sam Rose posted a chart to X. It showed Stack Overflow question volume over time: a slow climb from 2008, a peak around 2014, a gradual decline through 2019, then a cliff edge after November 2022. The last data point sat at a level last seen in 2008.
Fifteen years of growth. Gone.
The conversation that followed focused on the wrong thing. People debated whether Stack Overflow was dead, whether AI was better, whether the moderators had been too harsh. Nobody asked the structural question the chart was actually answering.
What Stack Overflow actually was
Stack Overflow wasn't a Q&A site. That's like calling Wikipedia a text file.
It was a knowledge accumulation machine. Every question asked, every answer written, every upvote cast, every accepted answer marked: permanent contributions to a shared technical commons. The 58 million questions in its database represent an enormous volume of human cognitive work, transferred into searchable, linkable, citeable form. For fifteen years, every time a developer hit a problem and typed it out, that problem and its solution entered the permanent record.
That record is what trained the models now answering questions instead of Stack Overflow.
Here's the part nobody is writing about: the replacement doesn't contribute back.
When you ask ChatGPT how to parse a datetime in Python, you get an answer. It disappears. The conversation logs to a server somewhere, feeds a feedback model, maybe improves future responses. It doesn't create a permanent, indexed, community-validated entry in a public commons that anyone else can find, cite, or build on.
We replaced a system that accumulated with one that consumes.
The printing press had an exit ramp
In 1492, a Benedictine abbot named Johannes Trithemius wrote a treatise against the printing press. His argument: hand-copying sacred text was a spiritual discipline. Mechanical reproduction would make monks intellectually lazy. The knowledge would degrade.
He was wrong about the degradation. The printing press democratized production: more people could create, more texts survived, literacy expanded. But he wasn't entirely wrong about the monks.
The scribes of the 15th century were knowledge curators. They filtered, corrected, annotated. Some of them, like Peter Schoeffer, adapted and became printers. Others didn't. Vespasiano da Bisticci, described as "the most noteworthy of all manuscript bookdealers," simply closed his shop.
The critical difference between then and now: when printing replaced scribal culture, it replaced a curatorial function with a more distributed, more productive version of the same function. More people could write, copy, distribute. The commons expanded.
AI replacing Stack Overflow does the opposite. Fewer people contribute to any shared record. The cognitive work happens in a private conversation, then evaporates.
The model collapse nobody is talking about
Daron Acemoglu, in a February 2026 NBER working paper, models what happens when AI substitutes human effort in knowledge-intensive tasks. His term for the outcome: knowledge collapse. As agentic AI provides high-precision answers directly, human effort drops, the flow of new public knowledge weakens, and collective general knowledge degrades over time.
He's describing an economic model. But Stack Overflow is the empirical data.
The questions are gone. That means future problems, the ones with new frameworks, new runtime environments, new edge cases, have nowhere to go. When the next generation of models needs to answer a question about a library released in 2027, there will be no Stack Overflow thread to draw from. The training corpus is aging. The models will consume knowledge created before their own dominance, until that supply runs out.
The scribes wrote manuscripts that lasted centuries. The Stack Overflow answers will keep training models until the questions stop coming. That moment has already arrived.
The scribes who in 1476 attacked and destroyed a printing press in Paris were wrong about the direction of history. But they understood something was ending.
We didn't even notice.