The principal–agent problem is the economic friction that appears whenever one party — the principal — hires or empowers another — the agent — to act on its behalf, but the two have divergent interests and the principal cannot fully observe what the agent does. Because effort is hidden and incentives diverge, a self-interested agent will sometimes act against the principal's interest, and the resulting loss (plus everything spent trying to prevent it) is the cost of the relationship. It is one of the load-bearing ideas of modern economics, running from corporate governance to insurance to the design of every DAO.

The problem was formalised independently in 1973 by the economist Stephen Ross and the political scientist Barry Mitnick, and it sits at the centre of the theory of agency. Any organisation that separates ownership from control — shareholders from managers, citizens from officials, or token holders from the people who actually move a treasury — is a principal–agent relationship, and inherits its problems.

Two information frictions: moral hazard and adverse selection

Agency problems come in two flavours, distinguished by when the information gap bites.

• Moral hazard (hidden action). After the relationship begins, the principal cannot observe the agent's effort or choices — only a noisy outcome. A fund manager who takes excessive risk with someone else's capital, or a core team that under-delivers because no one can tell diligence from luck, is the canonical case. Holmström's (1979) informativeness principle is the classic result here: an optimal incentive contract should condition pay on every signal that carries information about the agent's action, and only on those signals.
• Adverse selection (hidden information). Before the relationship begins, the agent knows something the principal does not — their true competence, or the real quality of what they are offering. Akerlof's (1970) “market for lemons” showed how this can unravel a market entirely, as the parties who know they are low-quality are the keenest to transact.

Both frictions are symptoms of the same root: asymmetric information. Where a principal could perfectly observe and verify an agent's behaviour, the problem would vanish; the interesting economics is entirely about what to do when they cannot.

Agency costs: monitoring, bonding, and residual loss

Michael Jensen and William Meckling (1976) gave the problem its most durable accounting. The total agency cost of a relationship, they argued, is the sum of three parts:

• Monitoring costs borne by the principal — audits, dashboards, reporting requirements, oversight committees — to observe and constrain the agent.
• Bonding costs borne by the agent — posting collateral, accepting vesting or clawbacks, submitting to rules — to credibly signal they will behave.
• Residual loss — the welfare gap that survives even the optimal amount of monitoring and bonding, because eliminating the last increment of misalignment costs more than it is worth.

The framing matters because it kills the naive hope of a perfect fix: some residual loss is efficient. The design question is never “how do we make the agent perfectly faithful?” but “what mix of monitoring, bonding, and tolerated slack minimises the total?” Eisenhardt's (1989) review remains the standard map of how those trade-offs shift with risk, observability, and outcome uncertainty.

Standard remedies — and why none is complete

Every institution that manages other people's money or power leans on some combination of the same levers:

• Transparency and monitoring — make the agent's actions observable, which shrinks the hidden-action gap but never closes it and is itself costly.
• Incentive alignment — tie the agent's payoff to the principal's outcome (equity, performance pay, treasury exposure) so their interests converge.
• Bonding and skin in the game — require the agent to have something at stake that they lose by defecting.
• Reputation — repeated interaction lets a principal punish defection in future rounds, which disciplines an agent who values the relationship.
• Competition — a contestable market for the agent's role (managers, delegates, service providers) makes replacement a credible threat.
• Exit — Albert Hirschman's (1970) third lever: rather than voice a complaint, the principal simply leaves, taking their capital with them. A credible exit disciplines an agent more sharply than any complaint, because it is not up for negotiation.

None is a silver bullet. Contracts are incomplete — no clause anticipates every future state — and every remedy has a price, which is exactly why transaction costs and residual loss never fall to zero.

The principal–agent problem in DAOs

A DAO does not escape agency by being on-chain; it relocates it. Token holders are the principals, but they rarely execute decisions themselves. The agents are everywhere:

• Delegates who vote on behalf of thousands of dormant holders in token-weighted systems;
• Core teams and foundations that set the roadmap and control the multisig;
• Multisig signers who hold the keys to the treasury between votes;
• Service providers and treasury managers paid to allocate the DAO's capital.

The on-chain twist is that monitoring is unusually cheap — every transaction and vote is public — while the other remedies are unusually weak. Delegates typically post no bond, competition for their seats is thin, and exit for a dissenting holder usually means dumping the token into whatever market exists while the contested treasury stays behind. Accountability tooling such as Karma and DeepDAO exists precisely to restore the monitoring and reputation levers that off-chain institutions take for granted. Nearly every entry in the catalogue of DAO failures — captured multisigs, rubber-stamp delegates, treasuries drained by insiders — is an agency problem wearing a governance costume.

Why token holders delegate anyway: rational apathy

The deepest reason agency persists in DAOs is that monitoring is rational to skip. A holder with a small stake bears the full cost of researching a proposal but captures only a sliver of the benefit of voting well — the same collective-action logic that produces rational voter apathy in any large electorate. So most holders delegate or abstain, control concentrates among the few who show up, and the gap between the principals' interests and the agents' actions widens. Vitalik Buterin's critique of coin-voting governance is largely a catalogue of the agency failures this apathy invites. The design response is to make participation cheaper or more rewarding than passivity — or to give the disengaged principal a lever that does not require them to monitor at all.

How Caper approaches this

Every caper inherits the principal–agent problem the moment holders stop executing decisions themselves. Caper narrows the agency slack along two axes, both outcome-level rather than novel machinery:

• A standing exit, not just a vote. The remedy Hirschman calls exit is usually unavailable to a DAO member. In a caper it is a first-class action: a member can call exit at any time to redeem a share of the treasury equal to their canonical vote weight w = (t·v) / (V·T) — governance holdings t times recorded votes v, over total vote supply V and circulating supply T (verified against compute_vote_weight and exit in contracts/logic/src/lib.rs). Because the payout is pro-rata and always open — no proposal to pass, no majority to win — an agent cannot strand a dissenting principal: they can walk with their share of the treasury the moment they lose trust. That is a credible outside option, the sharpest of the classic agency remedies, made structural.
• Weight that decays without participation. The same formula multiplies holdings by recorded votes: a holder who never votes has v = 0, and therefore zero weight and zero exit share, however large the bag. A large holding on its own cannot capture control; influence and the treasury claim both accrue to participation, which pushes back on the rational-apathy drift that hands DAOs to whoever bothers to show up.

Neither eliminates the residual loss Jensen and Meckling described — nothing does — but both are aimed at the levers on-chain governance usually leaves weakest. See how proposals work for the settle-then-execute path an agent's decision must clear.