The Principal-Agent Problem: Why Without Understanding This, Organizations Always Seem Like Palace Dramas

There’s a type of person in life who could be described as having an “easily deceived nature.” When renovating, they’re coaxed into spending more; when getting a phone repaired, they’re told the motherboard is broken. If they become a leader, they’re effortlessly manipulated by subordinates, yet afterward, they might even feel the team atmosphere is excellent. Reflecting later, they say: “I just tend to think too well of others.”

The opposite type of person exists. Facing the same group of people and complex situations, they not only avoid being duped but also operate in a well-organized and decisive manner, with no one daring to mislead them.

What do you think is the difference between these two types of people? Is the second type more assertive? Or better at reading people? Or stricter in their management?

While common folks often resort to explanations of morality and personality, having studied so many thinking tools, you undoubtedly possess insights that go beyond superficial narratives. Instead of examining morals and human nature, it’s better to scrutinize responsibility, power, interests, and information structure.

In this lesson, we’ll discuss an economic concept called the “principal–agent problem.” It can explain a range of phenomena, from being “scammed” and “management failures” to “official corruption.” One could say that if modern people don’t understand this principle, they will forever view organizations like watching palace intrigue dramas.

Understanding the Principal-Agent Problem #

Simply put, the principal-agent problem refers to “problems that arise because the person doing the work for you is not you.”

You are the “principal,” meaning the one who wants something done; the one actually doing the work for you is the “agent.” The agent is not you, so what they want isn’t perfectly aligned with what you want: you want quality, they want profit; you want the long-term, they want this month’s bonus. But they are on-site, and you can only listen to reports, so they know more than you do. When interests diverge and information is asymmetrical, problems arise.

As a shareholder, you are the principal, and the CEO, as the agent, manages the company for you—you care more about stock price, while they might care more about their salary and cashing out. You are the patient, and the doctor makes judgments for you—an extra test ordered means extra income for them. You are the voter, and officials manage public affairs for you—you, the nominal principal, are precisely the person furthest from power and with the least information…

Note that this has nothing to do with whether people are “good” or “bad.” The problems between you are purely determined by the positional structure of “principal-agent.”

People’s intuition is, why not just find a good person to be the agent, someone loyal, one of “us”? Yet Han Fei, over two millennia ago, already understood: “Superiors and subordinates engage in a hundred battles a day.” [1] How can there be a family between ruler and subject? They secretly contend a hundred times a day. He also said: “The interests of ruler and subject differ, hence no subject is loyal.” If interests aren’t aligned, how can loyalty be reliable?

Han Fei’s prescription for the principal-agent problem was “shu” (术), the art of statecraft, which is, in plain terms, a set of covert control techniques. The ruler must conceal their true intentions, use trusted confidants to spy on each other, and release false information to test subordinates… Shu is not institutionalized trust, but personalized anti-deception, which one could also call palace intrigue. We modern people must have better solutions.

Three Major Roots of the Principal-Agent Problem #

There are three fundamental mechanisms behind the principal-agent problem, which you’ll find familiar as soon as you hear them.

One occurs before the delegation: you can’t clearly see what kind of person the prospective agent is. This situation is what we’ve discussed as a “lemon market”—good and bad options look similar, and under deceptive influence, you often pick the most unreliable person. This is known as “adverse selection.”

The second occurs after delegation: you can’t clearly see what they’ve done. If you can’t see, it’s hard to hold them accountable; if it’s hard to hold them accountable, they act recklessly. This is called “moral hazard,” a situation we touched upon when discussing “soft budget constraints.” If you’re not on-site and can only rely on reports, if they hide something once and you don’t notice, they’ll become bolder over time. In Han Fei’s words, this is “subordinates conceal their private interests to test their superiors.”

The third mechanism was proposed by two Nobel laureates in Economics, Bengt Holmström and Paul Milgrom, and is called the “multitask principal–agent problem.” [2]

The logic here is that your delegation usually isn’t just one task, but a set of tasks—some are easy to measure, while others are extremely difficult to measure. You naturally reward the easy-to-measure tasks, so the agent will divert effort from the hard-to-measure tasks to focus solely on the easy-to-measure ones.

For instance, during your home renovation, you tell the contractor: “The sooner the project finishes, the better; an early completion bonus will be given.” So, they will rush the schedule fiercely. In places you can’t see, like how many coats of waterproof paint are applied or if the concrete cures for enough days, they will cut corners. Three years later, your bathroom leaks, and you curse the renovation team for being unethical, but wasn’t the reward function originally set by you?

Indeed, this is also a manifestation of “Goodhart’s Law.” If the education bureau only assesses university admission rates, schools will focus on rote learning and won’t care about students’ curiosity—even if the education bureau also considers curiosity very important.

Agents are not saints. With these three mechanisms at play, it’s probably normal for problems to arise.

Goal Translation: The Deeper Reason for Delegation #

Behind these three mechanisms lies a deeper reason, which we might call the “Goal Translation” problem.

Delegation isn’t just throwing out a task; it’s translating your goal into a structure that others can execute. And translation, by its nature, inevitably involves some loss.

For renovation, what you originally wanted was “a safe, comfortable home that you can live in for twenty years without headaches.” But how do you express this goal? Your contract states: 200,000 yuan, ninety days, including these specific items.

Then the renovation contractor must translate this down another layer for the师傅 (master craftsmen) responsible for execution. Along this entire chain—from value goal → contract terms → performance indicators → rewards and punishments → agent’s actions—the goal narrows a step with each translation. Every time the task is re-delegated, decision rights, accountability, accurate information, and incentives/penalties become misaligned.

Power without responsibility leads to impulsive decisions; responsibility without power makes one a scapegoat; knowledge without power means frontline information cannot prevent organizational blunders; benefit without responsibility leads to freeloading, gaming metrics, and cutting and running.

Our previous discussion on “commander’s intent” was about ensuring subordinates fully understand the superior’s true goals. Solving the principal-agent problem then asks a further question: after understanding, why should they be willing to act according to that goal, and why can’t they surreptitiously swap out the goal for another?

Three Core Rights a Principal Must Uphold #

“Theatrical thinking” (戏曲思维), also Emperor Chongzhen’s method for solving principal-agent problems, was “trust people you use, don’t use people you doubt.” I’ll gamble on your character! If I trust you, I’ll trust you fully, and I’ll listen to everything you say! But if I ever find out you’ve deceived me, I’ll kill you without hesitation.

The essence of modern delegation, by contrast, is precisely “exercise doubt when delegating”—but don’t use suspicion to police individuals; use it to design the structure.

Simply put, there are three things a principal must never relinquish:

First is the “right to interpret goals”: the agent works for you, but they cannot unilaterally define what “is good for you.”

Second is the “right to interpret information”: the agent reports to you, but they cannot be your sole source of information and interpreter of reality.

Third is the “final judgment right”: the agent, as the executor, cannot evaluate their own work.

This is not easy, and you too must take action. As a principal, you don’t have to personally do everything, but you must learn to set requirements, observe the process, ask for evidence, and verify results.

The right to interpret goals requires you to clearly think about what you truly want. Recall the three elements of “commander’s intent”: why, what counts as completion, and what cannot be sacrificed.

The right to interpret information requires you to verify facts independently of reports. Holmström also proposed the “informativeness principle”: the only valuable signal is one that helps you separate “how well he performed” from “how well he talked.” [3]

For this, you need information channels that bypass the agent. This could be original records—receipts, photos, system logs; it could be interim inspections—don’t wait until the last day for acceptance; or it could be third-party verification—audits, home inspectors, peer reviews. Supervision isn’t a lack of trust; supervision is the infrastructure of trust.

The final judgment right requires you to define “what counts as completion” before work begins and to tie payments to milestones. Acceptance criteria must be in black and white beforehand, and payments must follow the agreed-upon stages.

The final judgment right also includes retaining “residual judgment rights”: this borrows from the economic concept of “residual control rights,” meaning that for anything not covered in the contract, someone must ultimately have the final say. [4] Any expenditure beyond the budget must be approved by you; any change to the core plan must be approved by you. At the same time, you must always retain the ability to replace the person. If an agent knows you cannot do without them, they will slowly become your controller.

“Trust people you use” sounds magnanimous but is, in fact, a dereliction of duty. It doesn’t mean you have to micromanage people, but you need to align the structure of goals, information, and judgment rights. Once the structure is sound, you won’t need to be paranoid daily, and they won’t need to constantly second-guess your mood; both parties will feel more secure and at ease.

The Essential Difference Between Modern Institutions and Han Fei’s Statecraft #

Do you see it? The biggest difference between the solutions we discuss and Han Fei’s “shu” is that everything we propose is in the open, while “shu” operates in the shadows. Han Fei’s original words are clear: “There is nothing better than overt law, yet tactics should not be seen.” The key to shu is secrecy—to hide the measuring stick and make people guess. You can’t fathom my intentions, nor can you fathom my standards for reward and punishment, so you can only fear me. As Kong Yingda of the Tang Dynasty said: “If punishments are unknown and authority inscrutable, the people will fear their superiors.” [5]

The Qing Dynasty, which brought imperial rule to its pinnacle in China, invented a highly sophisticated form of shu: secret memorials. Emperor Kangxi allowed about a hundred trusted confidantes to bypass all government offices and write directly to the emperor. Yongzheng expanded this list to over 1,100 people, with even some mid- and lower-level officials in prefectures and counties gaining the right to submit secret memorials directly to the throne. [6] The cleverness of this shu lay in the fact that who held the right to submit secret memorials was itself a secret—every governor-general might have another pair of the emperor’s eyes nearby. You might report auspicious omens in your memorial, while the surveillance commissioner’s secret memorial might be detailing disaster. This is precisely how to prevent anyone from becoming the sole source of information.

The result was that tens of thousands of secret memorials all converged on one person’s desk, meaning the entire empire had only one pair of eyes for auditing. Yongzheng, during his thirteen-year reign, had to approve dozens of memorials daily, making himself the most overworked emperor in history.

Meanwhile, we modern people have contracts, arbitration, and courts. Modern auditing is an industry, a set of standards, involving thousands of trained eyes. Modern “secret memorials” are the public ratings given by tens of thousands of strangers to renovation companies.

However, this open system has one requirement that Emperor Yongzheng would never have accepted: it constrains not only the agents but also the principals.

Principal’s Self-Restraint and Agency Costs #

If you can arbitrarily change rules, withhold final payments, make retrospective demands, and renege on agreements, the agent will certainly take countermeasures in advance—they will raise their prices, withhold information, and save some energy for finding alternatives. Modern delegation doesn’t speak of “loyalty”; it speaks of a space for cooperation: you must make them believe that doing things well according to the rules is more cost-effective than anticipating your moves, circumventing you, or wearing you down.

So you must bind yourself in at least three areas.

First is to provide fair compensation. If you push the price so low that the agent has no legitimate profit, they will inevitably seek “hidden compensation” by cutting corners.

Second, allow the agent to say “that’s impossible.” If you say, “I want it fast, cheap, high-quality, and without risk,” and don’t allow anyone to say it can’t be done, you are forcing people to lie.

Third is not to punish bad luck as culpability. Results involve both effort and luck. If you attribute all bad outcomes to the agent’s fault, they will only pursue safe but mediocre tasks and, incidentally, learn to conceal bad news. A mature principal does not punish bad news—they punish the concealment of bad news.

Good delegation is never free. Economists Michael Jensen and William Meckling long ago clarified the accounting: solving the principal-agent problem requires paying three types of “agency costs”: [7] one is monitoring costs, the money you spend to guard against them—checking progress, reviewing records, hiring third parties, all consuming your time and money; the second is bonding costs, the money they spend to prove their innocence—paying deposits, providing warranties, accepting audits, a cost often factored into their quote and ultimately borne by you; the third is residual loss, the part where, even after the first two costs are spent, their actions and your interests still don’t perfectly align… Theoretically, the cost of supervising deviations down to zero is infinite, so deviations will always exist.

So, where is such a good thing as “trust people you use”? That, in essence, is refusing to pay the cost of delegation.

The good news is that agents are not solely focused on calculating against you. A counterpart to agency theory is “stewardship theory,” which reminds us: agents can also be driven by professional honor, a sense of mission, and your respect for them. [8]

If you can guard against unscrupulous individuals through systems, you can treat people as gentlemen in your relationships.

Renovation and Recruitment: Practical Examples of Modern Delegation #

Having discussed the core principles, let’s look at two application scenarios.

First, renovation. Poor delegation is a single sentence: “Budget 200,000 yuan, please make it nice for me.” The goal is undefined, all information is in the other party’s hands, and acceptance relies solely on a final glance. This isn’t delegation; it’s a wish.

Good delegation looks like this: “Budget within 200,000 yuan. Plumbing, waterproofing, and environmentally friendly materials are non-negotiable. Photos must be taken and archived at each stage, and material brands and models verified upon arrival. Payment will be in four installments, with the next payment only released after the waterproofing test is passed. Any changes to major materials must be confirmed with my signature. 10% final payment will be retained, to be settled one month after move-in if no major issues arise.”

The first two sentences define the right to interpret goals. The middle part defines the right to interpret information: archiving, verifying models, observing stage by stage, transforming invisible work into visible milestones. The latter part defines the final judgment right: money follows acceptance, major changes require approval, and that retained final payment is your ultimate guarantee.

Notice that this delegation statement contains no moral exhortation, no attempt to guess intentions. It doesn’t require the agent to be a saint; it simply requires the agent to be a normal person: within this structure, doing a good job is their most cost-effective choice. This is what we call “incentive compatibility.”

Another example: hiring a key employee responsible for sales growth.

Poor delegation is grandiose: “You’re in charge of growth, let results speak for themselves, I trust people I use!” In that scenario, the quickest way for the agent to boost numbers is naturally to buy traffic, offer subsidies, and overdraw on the brand… Two years later, they might jump ship with the data and channels.

Good delegation should be like this: “You’re responsible for growth, but user retention, brand reputation, and compliance risks are non-negotiable. Bonuses will be determined by new revenue, retention rate, complaint rate, and team collaboration. A portion will be paid in the current year, and another portion based on the second year’s retention. Large-scale advertising must undergo post-mortems. All important data must be entered into the company system, not just in your personal spreadsheets.”

A little secret about recruitment: truly skilled professionals won’t be scared away by such delegation. On the contrary, they will breathe a sigh of relief. Clear rules mean they don’t have to guess your intentions, don’t have to guard against you going back on your word later, and their contributions are clearly calculable and secure.

Vague delegation drives away gentlemen and attracts gamblers.

Conclusion: Modern Wisdom Beyond Palace Intrigue #

Han Fei saw the principal-agent problem earlier than anyone, but Han Fei lived in a world with only one principal: the emperor, the sole client, who had to guard against everyone. Therefore, Han Fei’s answer could only be shu. Shu is something ingrained in an individual; it cannot be universally applied or inherited: Yongzheng’s descendants might not possess his immense energy and cunning.

In modern society, everyone is a principal. Therefore, the modern answer must be scalable, relying on a design that anyone can learn.

Recent TV shows love to depict palace intrigue, making everyone think that modern offices are daily reenactments of “Empresses in the Palace”… Little do they know that palace intrigue is not China’s destiny.

We possess modern institutional conditions that Han Fei and Yongzheng lacked.

But “trusting institutions” doesn’t mean “if something goes wrong, there’s always the law”—institutions are something you must actively operate: a clearly written contract, an interim milestone, an unpaid final installment, an information channel that bypasses the agent.

This world isn’t necessarily made better simply by being sincere, but it’s also far from being so bad that deceit and manipulation are everywhere—it can always be designed for the better.

A verse bears witness:

Between high and low, a hundred battles deep, Vermilion edicts, an emperor’s spirit couldn’t keep. Yet with three rights, like stable railings, found, Delegation proceeds, no need for doubt to confound.

Notes #

[1] Han Feizi, Yang Quan: “The Yellow Emperor said: ‘Between high and low, a hundred battles a day.’ Subordinates conceal their private interests to test their superiors; superiors wield measures to divide their subordinates.” Han Feizi, Nei Chu Shuo Xia Liu Wei: “The interests of ruler and subject differ, hence no subject is loyal, and thus subjects’ interests are established while the ruler’s are diminished.” Also refer to Han Feizi, Nan San: “Laws are best made public, but tactics should not be seen,” Han Feizi, Ding Fa on the distinction between “law” and “shu,” and Han Feizi, Zhu Dao: “The Dao lies in the unseen, its function in the unknowable.”

[2] Holmström, Bengt, and Paul Milgrom. “Multitask Principal-Agent Analyses: Incentive Contracts, Asset Ownership, and Job Design.” Journal of Law, Economics, & Organization 7, special issue (1991): 24–52.

[3] Holmström, Bengt. “Moral Hazard and Observability.” Bell Journal of Economics 10, no. 1 (1979): 74–91.

[4] Grossman, Sanford J., and Oliver D. Hart. “The Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration.” Journal of Political Economy 94, no. 4 (1986): 691–719.

[5] Chun Qiu Zuo Zhuan Zheng Yi, Annotation by Kong Yingda, Duke Zhao, Year 6.

[6] Yongzheng Chao Jin Zhe Ren Yuan Kao (Research on Personnel Submitting Memorials during the Yongzheng Reign), Zhonghua Wenshi Net (sponsored by the National Qing History Compilation Committee), http://www.historychina.net/zz/354456.shtml; Yang Qiqiao: Yongzheng Di Ji Qi Mi Zhe Zhi Du Yan Jiu (Research on Emperor Yongzheng and His Secret Memorial System).

[7] Jensen, Michael C., and William H. Meckling. “Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure.” Journal of Financial Economics 3, no. 4 (1976): 305–360. https://doi.org/10.1016/0304-405X(76)90026-X.

[8] Donaldson, Lex, and James H. Davis. “Stewardship Theory or Agency Theory: CEO Governance and Shareholder Returns.” Australian Journal of Management 16, no. 1 (1991): 49–64.