Five sketches of what community-owned coordination models could look like. Each follows the same pattern: the group owns the model, trains it on its own behavior, and decides what it’s allowed to do. None of these fully exist yet.
A research community trains a model on its own papers, issue threads, and conference Q&A. The model surfaces dormant overlaps between labs working adjacent problems, suggesting conversations that haven’t happened yet. Nobody surrenders IP to a central platform.
A cohort of builders at an accelerator runs a coordination model tuned on its own activity: what’s getting built, what’s blocked, what could combine. The cohort owns the model while it’s together, and passes it on or dissolves it when the cohort ends. This is Batch 001.
A mutual aid network runs a model trained on its members’ own requests and offers. The model suggests matches, flags patterns, notices who hasn’t been heard from this month. The network’s organizing committee decides what it optimizes for and what it’s allowed to do. Training data stays inside the network.
A neighborhood keeps a model of its own deliberative history: what issues came up at last year’s meetings, who argued what, which compromises held and which didn’t. New residents can ask it what the block has already tried. The community board holds the weights. The city doesn’t.
A diaspora trains a model on its cross-generational correspondence, recipes, oral histories, and community records. It helps members find each other across time zones and translate between generations. Practices that would otherwise be lost to distance stay inside the community, readable.
Most of these don’t exist in any real form yet. The practices are ad hoc, the infrastructure is expensive, and the training recipes aren’t public. This co-op is a first instance, and everything we build gets published openly: models, code, training techniques, and the insights that come out of the process. The next group has less to invent.