When authorizing an agent to execute an operation, two fundamentally different architectural choices exist: single-agent authorization lets one agent independently hold complete authority to sign a transaction, with any operation executing once that one agent alone agrees; multisig authorization requires a certain threshold number of multiple independent agents (or a combination of agents and humans) to all agree before an operation executes. The two architectures have a fundamentally different actual consequence when a single link goes wrong.