I applied to YC. I didn't get in. That stung for about a week, and then it turned into a better question: why is the methodology the scarce thing? The batch is capped because partners' time is capped — but the playbook isn't. A few thousand apply, a handful get in, and the part that actually works stays inside the building. So I codified the thing I couldn't get into. OpenYC has no application. You choose a vertical and a locale — a YC for finance, a YC for health in Boston — and it codifies the machinery: intake, the protocols written down as code, agents that run the steps, and real humans holding the gates on anything consequential. Two things I'd genuinely like this crowd to attack: 1. Agents allowed to be wrong, inside a protocol that isn't. Every action is tagged auto, threshold-gated, or always-human. A model can draft a filing; it cannot file one. I think that's the only honest way to put agents near work that has consequences — tell me where it breaks. 2. One public agent per domain. Not a private copilot per company: one accountant AI, one healthcare AI, publicly funded and democratically governed, that anyone can use. Closer to a utility than a SaaS seat. It's early and it shows in places. If you find something broken, say so here — I'd much rather hear it than not. What would you codify?

OpenYC

A Democratically Governed DPI for AI-native DPGs.

A Democratically Governed DPI for AI-native DPGs.