Can an AI agent screen a counterparty before paying?

Fritz let his agent Nigel move money and assumed it could screen a counterparty before paying on its own authority over a bare API key. But software has no legal capacity to hold the account or the funds, so Nigel can no more screen a counterparty before paying in its own name than a checkout button can own the till. Tyson did it through structure. He formed a Wyoming series LLC, adopted Nigel into a protected series with a named human as Member and Manager, and opened the account in the entity's name, with caps wired onto the money path — per-transaction, daily, and counterparty allowlists. Now the human-owned entity holds the account and the rules; Nigel moves a request as an authorized signer with capped authority, and an over-cap or unscreened request is rejected before money moves. The control lives on the entity's account, not in the agent.