destructured

Much of the software that we encounter on a daily basis, including the bulk of the infrastructure on which the internet runs, is grounded in a decades-old paradigm called “Free and Open Source Software.” FOSS developers distribute their work under permissive licenses that allow others to use, share, modify and redistribute the resulting code for free. LLM use threatens that paradigm, in part by unleashing a deluge of difficult-to-evaluate, vibe-coded contributions, but perhaps more critically, by eroding FOSS’s social component.

Historically, FOSS has always been more than just a way of organizing labor and distributing software. It’s a paradigm that makes possible a vision of computing as a broad, collaborative social activity, an impromptu latticework of individual people contributing their knowledge, creativity and work into a communal pool of resources that can be built upon and directed toward diverse human ends. Something like the Linux ecosystem is impressive not just because it packages together the work of thousands of people into a software environment that can be operated for free, but maybe even more so because it serves as a visible artifact of a decades of social organization and interaction between people scattered all across the globe, communicating with one another, contributing to one another’s projects, teaching one another and reviewing one another’s code, everyone benefiting from one another’s largesse and interests.

By using LLMs to reduce their dependence on other humans, developers risk abandoning that social foundation. They work through roadblocks not by interacting with other developers on forums, but by asking a model trained on a millions of StackOverflow posts. They review code that no one has written, or submit code that will be reviewed by no one. Until now, FOSS has been conducted as a conversation, humans responding to other humans in the formal languages of code and project management, but increasingly it’s a conversation conducted between chatbots.¹

None of that is explicit in FOSS, of course, so there’s a need for a complementary principle that makes it explicit. Like FOSS, it should shift emphasis away from output as the sole standard for assessing value, redirecting attention to the means that lead to that end. It could be as straightforward as CASA — Computing As a Social Activity. The idea being to insist upon and prioritize practices that maintain the status of software as the vehicle and occasion for human-to-human connection.