Stack Overflow Co-Founder Issues Stark Warning to AI Industry as He Mourns Father's Passing

Breaking: Jeff Atwood Warns AI Companies Not to Destroy the Human Communities Behind Their Training Data

Jeff Atwood, co-founder of Stack Overflow and Discourse, has issued a dire warning to the AI industry: do not kill the communities that produce the training data your models depend on. In a personal blog post reflecting on the recent death of his father, Atwood emphasized that large language models (LLMs) owe their coding abilities to the high-quality Creative Commons dataset built by millions of Stack Overflow contributors.

"If the LLMs end up hollowing out the very communities that produce all their training data, they're going to really, really regret that," Atwood wrote. He urged AI companies to "treat the community with the respect they deserve... that we all deserve."

Key Quote from Atwood

"I’ll give these LLM / GAI companies the same advice I gave Joel Spolsky when I left Stack Overflow to start Discourse – do not, for any reason, under any circumstances, kill the goose that lays the golden eggs, aka the human community around your product that does all the real work."

Background

Jeff Atwood co-founded Stack Overflow, the world’s largest programming Q&A platform, in 2008. Its community-generated dataset is widely recognized as a critical resource for training coding-focused LLMs. In 2012, Atwood left Stack Overflow to start Discourse, an open-source discussion platform. He also recently reordered the Guaranteed Minimum Income (GMI) rural study counties so that Mercer County, West Virginia—his father’s home—would go first in October 2025. This trip was the last time he saw his father alive. "There is no loss, because nothing ever ends," Atwood said.

What This Means

The warning comes as AI companies increasingly rely on user-generated content for training, often without clear compensation or sustainability plans for the source communities. Atwood’s remarks highlight a growing tension: if platforms lose their human contributors, the quality of training data—and by extension the AI models—could deteriorate. "It’s pretty simple," he concluded. "Just treat the community with the respect they deserve."