Long before software, institutions solved delegation with paperwork. Power of attorney, mandates, spending limits — all crude but durable ways of letting someone act on your behalf without surrendering everything. The principal-agent problem was never eliminated; it was bounded.
Machine autonomy revives this old problem in a harsher form. An AI agent acting with your keys is not an employee you can interrogate or a fiduciary you can sue. It is a process. The only meaningful controls are the ones enforced before execution, because remedies after execution barely exist. Delegation, in other words, must become a technical primitive rather than a legal one.
Newton Protocol is worth studying precisely at this boundary. Structuring agent activity inside a rollup where permissions are explicit and enforceable turns delegation into protocol state rather than operational trust. That bounds certain principal-agent risks. But it also transfers the burden onto whoever specifies those permissions. A poorly scoped mandate enforced perfectly is arguably worse than a broad mandate supervised carefully, because the system will execute the mistake with total fidelity.
Centuries of institutional design assumed a human somewhere could exercise judgment when rules produced absurd outcomes.
If enforcement becomes perfect and judgment becomes optional, who absorbs the cost when a correctly executed mandate turns out to be the wrong one?
$NEWT
@NewtonProtocol
#newt
Machine autonomy revives this old problem in a harsher form. An AI agent acting with your keys is not an employee you can interrogate or a fiduciary you can sue. It is a process. The only meaningful controls are the ones enforced before execution, because remedies after execution barely exist. Delegation, in other words, must become a technical primitive rather than a legal one.
Newton Protocol is worth studying precisely at this boundary. Structuring agent activity inside a rollup where permissions are explicit and enforceable turns delegation into protocol state rather than operational trust. That bounds certain principal-agent risks. But it also transfers the burden onto whoever specifies those permissions. A poorly scoped mandate enforced perfectly is arguably worse than a broad mandate supervised carefully, because the system will execute the mistake with total fidelity.
Centuries of institutional design assumed a human somewhere could exercise judgment when rules produced absurd outcomes.
If enforcement becomes perfect and judgment becomes optional, who absorbs the cost when a correctly executed mandate turns out to be the wrong one?
$NEWT
@NewtonProtocol
#newt