Principal-Agent Problem

The principal-agent problem arises whenever one party, the principal, delegates a task to another, the agent, whose interests do not fully align with theirs, and who knows more about their own effort than the principal can observe. A shareholder hiring a manager, a patient trusting a surgeon, a voter electing a representative — each pairing has the same shape: the principal wants an outcome, the agent controls the means, and the gap between what the agent does and what the principal can see is where interests diverge. Unmonitored divergence is itself a kind of Externality: the cost of the agent's shirking lands on the principal, not the agent who caused it.

Monitoring narrows that gap but never closes it, and it is never free — auditing a manager or reviewing a contractor's hours costs time and money that could otherwise go to the task itself. Incentive contracts try a different route: tie the agent's reward to an observable proxy for the outcome the principal actually wants — commission for revenue, a bonus for a target, a grade for learning. That requires the outcome to be made legible first, in the sense at Legibility, and that is where the arrangement meets Goodhart's Law: the proxy is chosen because it is measurable, the agent optimises the proxy because that is what they are paid for, and the correlation between proxy and true outcome erodes as the incentive succeeds at moving the proxy.

Every fix compounds the problem rather than resolving it. Add a second metric to catch gaming of the first, and the agent optimises a weighted combination of two proxies instead of the outcome; add oversight to catch that, and the oversight becomes a target to satisfy rather than a check to pass. Regulatory Capture and Rent-Seeking are both principal-agent failures at the scale of a state — the public is the principal, the agency or license-holder is the agent, and the information the principal needs to monitor well is held by the party being monitored.