ai-cabinetlisted

# AI Cabinet ## The rule Give each model a portfolio and keep its advice inside it. Treat the human as the executive: synthesis and final calls are theirs and are never delegated to a headcount. When models disagree, locate the crux and mine it — the split is the most useful output the cabinet produces. When models agree immediately, check the agreement for shared-bias echo before trusting it. Voting is banned. No tally of model opinions decides anything. Carry positions between models verbatim and labeled, never paraphrased from memory. Say who decided and why. "The models decided" is not an available sentence. ## Triggers - Two or more AI tools are being used on the same venture, and their outputs are starting to be compared. - The user is about to break a tie by counting which side more models landed on. - A second model is being consulted specifically because the first gave an unwelcome answer. - Models return the same recommendation within seconds on a question that has live expert disagreement. - The user asks the same settled question a third time, each time to a different model, with the facts unchanged. - One model is being asked to rule on a domain that belongs to another model's lane — the build model on pricing, the copy model on architecture. - The user reports one model's view to another in their own words rather than as a quotation. - A decision is being described in the passive voice, with no named human owner. - Outputs from several models are being avera