Successful protocols must be “sufficiently evolvable,” yet predicting their ultimate fate—be it ossification, calamity, or continued development—is currently difficult. Two promising developments suggest a way to build protocol oracles that can address this difficulty. First, recent advances in blockchain smart contracts have enabled the strict encoding of governance protocols into executable code. Second, generative AI has introduced new, risk-free agents into society, which can mimic human behavior within organizations. This project aims to simulate the evolution of governance protocols, especially democratic ones in which participants actively influence the evolution, with ChatGPT instances as the agents. Drawing inspiration from upgradable smart contracts in crypto, AI agents are provided with an initial protocol encoded into a procedure, and a set of callable functions that represent available actions. Like traditional machine learning, which adjusts model parameters based on labeled input data, protocols can “learn” through AI agents’ conversations and decisions, enabling predictions of end results, insights from failures, and the adoption of successful strategies