Robinhood's AI agent trading launch is a real milestone. It's also highlights something the industry often gets wrong. The question isn't "should agents manage capital?" They should. It's "what authority model they operate under?" Giving an agent keys or broad wallet permissions is the wrong answer. You can only explain misbehavior rather than prevent it. What financial agents actually need: → Constrained execution: an agent only act within an explicitly defined scope. Not "we trust it won't" but "it architecturally cannot" → Revocation: the user can pull authority at any time, without needing the agent's cooperation → Auditability: every action is visible and attributable before and after it happens, not reconstructed from logs after something goes wrong This must live at the authority layer, in the smart contract, not in a compliance dashboard on top of unconstrained execution. The agent might move your capital into a low-yield fund. That's fine, that's what you authorized. But it should be architecturally impossible for it to send funds to an unapproved address, bridge to an unapproved destination, or act outside the scope you set. Agents need contracts, not keys.