How do you limit what an AI agent can spend?

FX-08 PROB-06 gate: INFO cast: Fritz, Tyson ๐ŸŸข assertable

Fritz handed his agent Alfie the company card with no limit and went on with his day. Nothing stopped Alfie from spending up to whatever the card allowed, because the only boundary was Fritz's trust โ€” and trust is not a control the rail can read. An agent has no judgment the law recognizes and no authority of its own; it acts only for whoever stands behind it.

Tyson wrote the boundary down. In the operating agreement he set Alfie's spending authority and named a human as Manager who holds it; Alfie acts as an authorized signer with capped authority, nothing more. Each attempted payment is checked against that cap at the rail โ€” within the cap it clears, over the cap it is rejected before money moves. The limit now lives in a document and in code, where it can be pointed to, instead of in the operator's optimism.

MoralA spending limit only counts when it's written into the entity's documents and enforced at the rail; authority you can't point to isn't a cap.

What this does not mean. This shows where the cap is written and where it bites. It is not a promise that the agent never overspends, that every rejection fires, or that the operator is protected โ€” the cap is a configured control, dependent on the formalities and on the rail actually enforcing it.

See also: FX-15 โ€” caps on the money path ยท FX-18 โ€” over-cap rejects before money moves ยท FX-11 โ€” what each formation document does.

W.S. ยง17-29-211 (Wyoming series LLC)