How do spending caps work for an AI agent?

FX-15 MONEY-01 gate: INFO cast: Fritz, Tyson 🟢 assertable

Fritz handed his agent Nigel the company card and said "pay what you need to." There was no ceiling and no list — any amount, to any counterparty, on Fritz's own card, in Fritz's own name. The only limit on the money path was whatever Fritz happened to notice afterward.

Tyson wrote the limits down first. In the operating agreement of his Wyoming series LLC, with a named human as Manager, he gave Nigel capped authority: a per-transaction limit, a daily total, and a counterparty allowlist of names the agent may pay. Nigel signs as an authorized signer, never as a person or company the law recognizes. Those caps don't live only on paper — they're checked at the payment rail, so a request over the per-transaction limit, past the daily total, or to a name off the allowlist is measured against the written rule before any money moves. The agent proposes; the rail checks the proposal against the cap.

MoralCaps belong where the money moves — written into the operating agreement and checked at the rail, not left to whatever a human notices after the fact.

What this does not mean. This shows where caps are defined and where they are enforced. It is not a promise that Fritz was protected, that no money is ever lost, that any payment clears, or that the agent is safe to trust — a cap is a checked rule, not a guarantee, and separateness is performed through observed formalities, not secured by writing a limit down.

W.S. §17-29-211 (Wyoming series LLC)