Legislatures create agencies to carry out their will. Voters elect representatives to act on their behalf. Ministers direct civil servants. Shareholders hire executives. In each case, authority flows downward through a clear chain of command, at least on the organisational chart.

Yet the same pattern repeats across political systems. Agencies pursue priorities their creators never intended. Programmes are implemented in ways that surprise the officials who designed them. Delegated authority, once granted, seems to develop a life of its own. This is rarely a story of rogue bureaucrats or weak oversight. It happens in well-run systems with capable principals.

Political scientists call this the principal-agent problem, and its persistence is a puzzle. Why can't those who hold formal authority simply insist on compliance? The answer lies in three structural features of delegation itself. The first is who knows what. The second is what it costs to find out. The third is what no agreement can ever anticipate. Together, they explain why some loss of control is not a failure of delegation but a condition of it.

The first reason principals lose control is simple: agents are closer to the work. A regulatory agency inspecting factories learns about industrial processes, compliance patterns and enforcement trade-offs that no legislative committee can match. This is why delegation happens in the first place. Principals delegate precisely because agents can develop expertise the principal lacks.

That expertise creates two distinct forms of asymmetry. The first is hidden information. Agents know things about the problem, the options and their own capabilities that principals cannot easily verify. The second is hidden action. Principals can observe outcomes but rarely the effort, choices and judgements that produced them. When a crime rate falls or a programme underspends its budget, the principal sees a number. The agent knows the story behind it.

These asymmetries compound over time. Agencies develop specialised vocabularies, internal data systems and professional networks that principals cannot penetrate. They also control much of the information principals use to evaluate them. Budget justifications, performance reports and testimony before oversight bodies are all produced by the agent. A legislature trying to judge whether a defence procurement programme is on track will largely rely on the account given by the officials running it.

This is not mainly a matter of deception. Most agents report honestly. But honest reporting still involves selection: what to measure, what to emphasise, and what to leave out as mere context. Each choice shapes the principal's picture of reality. Charles Lindblom observed that policymakers never act with full knowledge. They act on partial maps drawn by others. In delegated systems, the agent usually draws the map. The principal's authority remains formally intact, but its exercise depends on information the agent supplies.

Takeaway

Delegation transfers expertise along with authority, and expertise is a form of power that cannot be handed back. Whoever produces the information a decision rests on shapes that decision, regardless of who formally makes it.

If information is the problem, monitoring seems the obvious solution. Principals can demand reports, conduct audits, hold hearings, appoint inspectors general and tie budgets to performance metrics. Every governance system uses some mix of these tools. The question is not whether monitoring works. It is whether it is worth what it costs.

Monitoring is expensive in ways that are easy to underestimate. Some costs are direct: staff, auditors and reporting systems. Others are opportunity costs. A committee spending months overseeing one agency is not legislating elsewhere. There are also performance costs. Agents under intense monitoring often shift effort from doing the work to documenting it. Teachers teach to the test. Hospitals manage waiting-list statistics. Police departments reclassify offences. The metric improves while the underlying goal may not.

Political scientists Mathew McCubbins and Thomas Schwartz described two broad strategies principals use. Police patrol oversight involves active, centralised scrutiny, which is comprehensive but costly. Fire alarm oversight relies on affected groups to complain when agents stray, so principals only step in when alerted. Legislators often prefer fire alarms because they are cheaper and politically rewarding. But fire alarms only ring when someone with resources and attention is harmed. Drift that affects diffuse or unorganised constituencies can continue unnoticed.

The rational outcome is a stable level of tolerated drift. Principals keep monitoring until watching more would cost more than the deviation they would prevent. Then they stop. This is not negligence. It is a reasonable allocation of scarce attention. But it means every delegated system contains a zone of discretion that principals have, in effect, chosen not to see. Who benefits from that zone depends heavily on which interests are organised enough to pull the alarm.

Takeaway

Every system of oversight contains a deliberate blind spot, sized by what principals can afford to watch. The interesting question is not whether agents drift, but whose interests decide where the alarms are placed.

Even with perfect information and free monitoring, a third problem would remain. Delegation agreements cannot specify what agents should do in every circumstance. This applies to statutes, mandates, contracts and job descriptions alike. The world generates more contingencies than any drafter can foresee. Economists call this contract incompleteness, and in political systems it is the rule rather than the exception.

Legislatures often write incompleteness in deliberately. Precise statutory language requires political agreement that coalitions may not have. Broad mandates allow legislators to claim credit for addressing a problem while passing the contested details to agencies. Examples include regulating in the public interest, ensuring safe workplaces and maintaining price stability. When those details later prove unpopular, the blame can be passed along too. Vagueness is frequently not a drafting failure but a negotiated settlement.

Whatever its origin, incompleteness hands agents residual control. This is authority over everything the agreement does not explicitly cover. When a new technology emerges, a novel financial instrument appears or an unforeseen crisis strikes, the agent interprets the mandate first. Central banks responding to financial crises have repeatedly stretched the meaning of their founding statutes. Environmental agencies have applied decades-old legislation to pollutants its drafters never imagined. Each interpretation becomes precedent, and precedent accumulates into policy.

Principals can respond by rewriting the rules. But this requires assembling a new coalition, with all the friction that involves. Agents know this. Because correction is hard, agents can safely adopt a wider range of interpretations. Over time, the gap between what was delegated and what is exercised grows. It grows not through defiance, but through a long series of reasonable gap-filling decisions, each defensible on its own terms.

Takeaway

Authority over the unanticipated is often more consequential than authority over what was written down. Those who interpret ambiguous mandates in moments of novelty quietly write the rules everyone else inherits.

Information asymmetry, monitoring costs and contract incompleteness are not separate flaws. They reinforce one another. Agents fill gaps in the mandate using knowledge principals lack, in areas principals cannot afford to watch.

This suggests a different way of reading political systems. The organisational chart shows where authority formally sits. But real power accumulates wherever information concentrates, oversight thins and mandates fall silent. Those are the places to look when policy outcomes diverge from stated intentions.

None of this makes delegation a mistake. Complex governance is impossible without it. Rather, control is always partial and continuously negotiated. It is shaped less by formal hierarchy than by who knows, who watches and who decides when the rules run out.