What We're No Longer Seeing: AI and the Invisible Newcomer in Open Source

Last winter, our team went through I-Corps training as part of the NSF's POSE—Pathways to Enable Open-Source Ecosystems—program. I-Corps is the National Science Foundation's customer-discovery training, and POSE adapts it for open-source projects. For us, that meant seven weeks and more than a hundred interviews with stakeholders in and around our ecosystem. (The "us" here is stdlib, an open-source standard library for JavaScript and Node.js, with an emphasis on numerical and scientific computing.)<br>These ecosystem-discovery interviews had one hard and fast rule: you don't pitch. No explaining what stdlib does, no defending decisions, no selling.<br>You ask, and you listen.<br>So that's what we did. The curriculum taught us how open-source ecosystems are supposed to be sustained—the frameworks, the canvases, the interview discipline. The interviews taught us something nobody had put on the syllabus. Talking with maintainers, contributors, and users, the same theme kept surfacing: AI was changing how people found projects, got help, and contributed. And we weren't just hearing it in the interviews. At conferences and in hallway conversations, in blog posts and LinkedIn threads, peers who do developer relations and community work for a living were describing the same shift from their own angles.<br>We went to learn about our ecosystem. While we were looking, the ecosystem was changing around us.<br>This post is about what we were seeing. It's also about what we weren't.<br>For a long time, open-source communities have depended on a specific kind of visible friction. People got stuck and they showed up—on Stack Overflow, in GitHub issues, on mailing lists, in forum threads. That first act of asking for help in public was itself a form of participation. It was the way you found the edge of someone else's community, and the way they found you.<br>There's an old model from Charles Vogl's The Art of Community—the idea that any community has concentric rings around it. People start as visitors. They find their footing. They ask, in one way or another, what am I doing here? The journey from the outer ring inward—visitor to member to elder—isn't automatic. It requires visible pathways. It requires people willing to guide. And, crucially, it requires that the path be known to others.<br>That model isn't unique to Vogl, or to open source. The figure below comes from a study of a community most readers have no skin in: recreational scuba divers in the early 1990s. Edouard Lagache, building on legitimate peripheral participation (a term coined by Jean Lave and Étienne Wenger), mapped the same geometry—old-timers at the center, casual sightseers at the outermost ring, a labeled trajectory pulling inward. People become members of a community by doing low-risk things at its edges—things that exist because there's a structure around them that makes them visible. Early 90s scuba diving is just an example, but the same dynamic and shape exists in open source, in sports, in fandoms, in any community you can name. The details differ, but the geometry is the same.<br>What made it work in open source, structurally speaking, was friction—visible friction. Someone got stuck, showed up, and both sides could see each other. That first public question was a form of initiation, and it was also legitimate peripheral participation in action: a low-risk, visible way of taking part from the edge. That dynamic is woven through decades of open-source community-building practice—it's underneath Ten simple rules for helping newcomers become contributors to open projects, among others. The public question made the newcomer visible, and it created the opportunity for someone already inside to respond.<br>It also made the community visible to the newcomer. Oh—there's a there here. There are people. There's a way in.<br>For a while, the shift we kept hearing about was a vibe more than a finding—something you could feel in the conversations before you could point to it on a chart. But it kept turning out to be tangible. At FOSDEM this year, David Allen, who leads developer relations at Grafana, showed everyone his numbers, describing a moment some maintainers haven't had yet but many will.<br>He'd gotten comfortable expecting community growth of 8 to 18 percent a quarter, almost automatically—I could fall asleep, he said, and the community would grow. Then one quarter the number dropped 27 percent. They checked the data, ran it again, and it wasn't a measurement error: the number of people entering the community had collapsed, even though overall user growth was still going up. The mechanism wasn't mysterious, either. Organic traffic to community spaces was down roughly 30 percent, traceable to question-and-answer behavior moving to LLMs.<br>And it's not a one-project story. A 2024 study in PNAS Nexus found a clean way to isolate the effect: compare Stack Overflow activity in English (where ChatGPT was...