Learn it → Apply it → Save as Skill
Whenever you hire, elect, or appoint someone to act on your behalf, their incentives and yours start to drift apart. The principal-agent problem names that drift: agents who understand more than principals can verify tend to serve their own interests to some degree, even when nobody breaks a rule.
The Dr. Wallace’s Favorite Mental Models is designed to turn useful ways of thinking into reusable tools. Each model gives you a fast visual overview, deeper interactive learning, practical application exercises and an AI-ready Markdown skill you can save for future use.
Instead of merely reading about mental models, you can gradually build your own AI skills library: a collection of portable reasoning skills that can help ChatGPT, Claude, Grok and other AI assistants apply the mental models you value.
A principal delegates a decision to an agent because the agent has time, skill, or information the principal lacks. That same gap, the reason the agent was hired in the first place, is what lets the agent's effort, judgment, or honesty drift away from the principal's interest without being caught right away.
Important: the gap is not proof of dishonesty. Even a fully honest agent optimizes for what is measured and rewarded, not for what the principal silently hopes for.
The model is most useful whenever you cannot watch every action an agent takes, but you still live with the consequences of what they decide.
Agency problems follow a repeatable shape once you know where to look for it.
A principal hands a decision or task to an agent because the agent has skill, time, or information the principal lacks.
The agent's own goals, workload, risk tolerance, or pay structure are not identical to the principal's. Interests separate, at least at the margin.
The principal can only partly observe the agent's effort or information. Full monitoring is usually expensive or impossible.
The principal responds with incentive alignment, monitoring, bonding, or some mix, accepting a residual cost as the price of delegation.
Open a use case to see what the model changes.
Set the measured target carefully. Reward only closed tickets and expect fast, shallow closes. Reward only revenue and expect discounting or corner-cutting elsewhere. Ask what a capable, self-interested employee would optimize for under the metric before you adopt it.
A fee tied to assets under management rewards gathering assets, not necessarily growing your money. A performance fee tied to gains rewards risk-taking with your capital. Check which side of that trade-off your fee structure actually sits on.
A typical commission gives an agent only a small personal upside for pushing your price higher. Waiting for a better offer costs the agent real time for a small personal gain, so many agents lean toward a quick sale over your top price.
A specialist paid per procedure has a built-in reason to recommend that procedure, even when watching and waiting is medically reasonable. This does not make the advice dishonest. It means a second opinion from someone without that specific incentive is worth seeking for major decisions.
Shareholders cannot watch daily management decisions, so boards lean on audited numbers, independent directors, and pay tied to long-run performance rather than short-run stock price, to narrow the gap between what management wants and what owners want.
Hourly billing rewards time spent, not the outcome delivered. A fixed price or milestone-based contract moves the incentive back toward finishing well and quickly. Ask what the payment structure rewards before assuming slow work reflects difficulty rather than the fee design.
This is an illustrative example designed to teach the method. It uses invented figures for a fictional seller.
Expand for the economists behind the formal theory and the evidence behind the real estate pattern above.
Economists Michael Jensen and William Meckling gave the problem its modern framework in their 1976 paper on the theory of the firm, describing the costs that arise whenever an owner (the principal) delegates decisions to a manager (the agent) whose interests are not identical to the owner's. Stephen Ross described a closely related structure slightly earlier, in a 1973 paper on the economic theory of agency.
The worked example above uses invented numbers, but it reflects a documented pattern. Economists Steven Levitt and Chad Syverson studied real estate transactions and found that agents kept their own homes on the market longer, roughly ten additional days on average, and sold them for a few percentage points more than comparable homes they sold on behalf of clients. The gap is consistent with an agent accepting a faster sale for a client while holding out longer when selling their own property.
Moral hazard is one specific mechanism inside agency problems: it describes how an agent's behavior can change once they are insulated from the consequences of their own actions, most often discussed in insurance and lending. The principal-agent problem is the broader structure, delegated authority plus imperfect alignment plus imperfect observation, that produces moral hazard as one of its symptoms, alongside adverse selection and simple shirking.
Most agency problems come from misaligned incentives, not dishonesty. A cooperative agent working exactly as paid can still drift from what the principal wants.
Full observation is usually too expensive to be worth it. Spend monitoring effort where the stakes and the information gap are both largest.
A measurable proxy, tickets closed, calls made, assets gathered, is not the same as the outcome the principal actually wants. Check what the proxy leaves out.
An agent who bears none of the downside of a bad outcome has little reason to avoid it. Some shared risk usually improves alignment.
Incentive design cannot fully substitute for choosing an agent whose own values already point in the principal's direction.
Incentives that worked when they were designed can drift out of alignment as circumstances change. Agency problems need revisiting, not a one-time fix.
the relationship runs on long-standing reputation and repeated interaction, where the cost of ever cheating already outweighs any one-time gain.
the agent's own goals and the principal's goals are already close, such as a founder running their own company or a volunteer serving a cause they chose.
the real problem is a skill gap, not an incentive gap. Better training helps more than a new contract when the agent is trying but unable.
This browser tool structures your thinking. It does not send your information anywhere or perform AI analysis.
Save as Skill converts the model into portable Markdown instructions that teach an AI when and how to apply the principal-agent problem, what questions to ask, what evidence to seek, what mistakes to avoid and how to structure its analysis.
Save the skills you find useful and organize them into your own reusable library. The Markdown is intentionally model-neutral, so it can be used wherever an AI system accepts uploaded files, project knowledge, custom instructions or reusable text context. Platform capabilities differ, so this is a portable skill file rather than a claim of one-click installation.
No. The same structure applies to time, effort, attention, and information, not only pay. A volunteer board member, an appointed regulator, or an elected representative can all diverge from the people they represent even with no direct financial incentive involved.
Usually not completely. Writing a contract that anticipates every situation is expensive or impossible, and a contract that is too rigid can push an agent to satisfy the letter of the agreement while ignoring its spirit. Most real solutions combine partial monitoring, partial incentive alignment, and some tolerance for residual loss.
Agency cost is the umbrella term economists use for the combined cost of a principal-agent relationship: what the principal spends monitoring the agent, what the agent spends demonstrating trustworthiness, known as bonding, and the value still lost even after both of those efforts, known as residual loss.