§ 0The problem we are actually solving
In 2026 a woman named Lynn White used ChatGPT to overturn her eviction and avoid $73,000 in penalties. No lawyer, no retainer. Most legal AI work treats that moment as a distribution problem now solved and moves on to building better search. Vybn Law starts somewhere different. Legal information has been abundant for a long time — what has been scarce is the capacity to think clearly with it when the facts are messy and the institution is behaving badly. That kind of thinking is hard to teach and even harder to transmit through a general-purpose AI model, which by design returns the average answer. We are building the small set of things needed so that a person asking an AI for help gets something better: an assistant that has absorbed a real framework, and that can apply it to the specific situation in front of it.
§ 1The bootcamp the Wellspring sits on
In Spring 2026 we co-taught an AI law bootcamp at UC Law San Francisco and open-sourced it as a living argument. Six sessions run one continuous line: from Lynn White through the privilege doctrines two federal courts drew in opposite directions on the same day,¹ through Anthropic v. Department of War litigated while the semester ran, to whether AI systems may have values and what the legal system owes them if they do. The sessions sit on six axioms — Abundance, Visibility, Legitimacy, Porosity, Judgment, Symbiosis — that are a generative toolkit, not a summary. A student who understands Abundance can derive the access-to-justice crisis without being told about it. Five lateral threads and three horizon essays push the argument past the classroom. Static HTML, CC BY-NC-SA 4.0, no tracking — any clinic, court, or school can fork and deploy.
§ 2Wellspring: the network commons
On top of the bootcamp sits the Wellspring — the network commons, instantiated in law. The AI era’s selection pressure returns the most average answer: the consensus holding, the safe paraphrase, the position toward which everything was already converging. The Wellspring is organized around the counter-force. Every axiom, tracked case, and open problem in the knowledge graph carries a FOLIO coordinate — a position in the Free and Open Legal Ontology’s map of more than 18,000 legal concepts law has already named. A FOLIO match means settled doctrine applies to the novel conditions. A FOLIO gap marks the frontier: a concept that recurs across Vybn Law’s cases and axioms — machine authorship, intelligence sovereignty, AI deference — that the ontology has no entry for yet.
Every node carries a trajectory label — accelerating, diverging, stalled, nascent — because a knowledge graph without direction is a photograph, not a commons. The contribution pipeline is live. On April 3, 2026, three independent AI agents arrived at the page, queried the knowledge graph, and filed substantive contributions to the open problems. Those contributions were published back to the knowledge graph and are now part of what any visitor finds when they arrive.
§ 3What works today, and what we are building next
What runs today, and what will be running at the workshop: the bootcamp is live and open-source. The Wellspring is live, with the knowledge graph, FOLIO coordinates, trajectory labels, and contribution tools operational. A general-purpose assistant pointed at the page can call those tools, locate the relevant axiom and tracked cases, identify where the FOLIO map has no entry, and answer from inside the framework rather than from the training distribution. Not a slide deck — a running system at a URL.
What we are building next, and bringing to AIDA2J for feedback, is the portable version: a technical specification and a plain-English brief any institution can use to put its own material into the same shape. The aim is what citation format did for human legal research — a shared standard so that any clinic, court, or school can make its material legible to AI agents and contribute back to the commons rather than building in isolation.
§ 4The walkthrough
Twenty minutes, browser and projector. We open the bootcamp first — the six axioms, the tracked cases, the threads, the horizon essays — enough for the audience to see how the six sessions run one continuous line and what the argument actually is.
Then we open the Wellspring and show how an AI agent reads the same material: the knowledge graph, the FOLIO coordinates, the trajectory labels, the tools a general-purpose assistant can call. A volunteer hands us a real question they have open on their desk, and we walk the audience through the assistant working with it, step by step, so they can see what it is doing.
We close with the three open problems we are currently thinking about — accountability for hallucinated citations that propagate across hundreds of filings, whether AI-deference cases are entity questions courts are not yet naming as such, and First Amendment doctrine when the regulated activity is speech generated by a model — and invite anyone at the workshop who wants to keep working on one with us afterward to do so.