First principles · The principal–agent problem
The Agent Without Appetite
Every institution you have ever stood inside is scaffolding around one old problem: anyone who acts for you has interests that are not yours. The first agent with none does not end the problem. It moves it.
The price of acting through someone else
M + B + R
Monitoring plus bonding plus residual loss: what you pay to watch your agent, what your agent pays to look watchable, and what leaks anyway. Jensen and Meckling’s 1976 invoice for the oldest overhead in economics.
The house went on the market Thursday. By Sunday there was an offer that felt low, and your agent — warm, competent, instantly likable — said take it. At the kitchen table that night you did the arithmetic she never would. Her share of another ten thousand dollars, after the brokerage split, was a dinner out. Her share of four more weekends of open houses was zero. You took the offer. You never did find out which of you the advice was for.
The oldest overhead
Adam Smith named the problem in 1776, watching the directors of joint-stock companies handle other people’s money. It cannot well be expected, he wrote, that they should watch over it “with the same anxious vigilance” that owners watch their own. “Negligence and profusion, therefore, must always prevail, more or less.”
Two centuries later, economists gave the problem a formal name — the principal–agent problem — and an invoice. Whenever you act through another, their interests leak into your outcome, and everything you do about that leak costs money. In 1976 Jensen and Meckling split the bill three ways: monitoring, what the principal pays to watch; bonding, what the agent pays to be watchable; and the residual loss, the leak that survives both.
Look around any institution and you are reading that invoice. The commission, which points the salesman’s hunger at your outcome, approximately. The audit. The license. The surety bond. The performance review. The middle layers of every org chart, which exist mostly to watch the layers below them. The tip. Half of civilization’s paperwork is receipts for M and B, purchased in the hope of shrinking R.
The residual loss is the honest term of the three. It admits that after everything you pay to watch, and everything they pay to seem watchable, some of your outcome still bends toward their interests — and that there is no price at which the bending stops. The famous realtor study measured the bend directly: agents selling their own homes leave them on the market longer, and sell them for more, than the homes they sell for clients. Same skill. Different appetite.
Figure 01
The invoice, itemized
Audits, reviews, dashboards, managers, inspections. The cost of watching someone whose interests are not yours.
Licenses, certifications, surety bonds, reputations. The cost of proving you are worth not watching.
The bend that survives both. Your outcome, tilted toward their interests, at any level of vigilance.
The agent without appetite
Now run the old machine on a new input. The synthetic delegate that sells your house, files your claim, negotiates your renewal — it has no quota ending this quarter. No rival client. No brokerage split. No fatigue at 4:50 on a Friday. No career in which your outcome is a stepping stone. For the first time in the two hundred and fifty years since Smith named the problem, you can act through something that wants nothing.
The obvious conclusion is that the agency cost goes to zero. The obvious conclusion is wrong.
The cost is not eliminated. It is conserved. The appetite you evicted from the kitchen table did not disappear. It changed floors.The claim of this piece
Interests did not leave the transaction. They moved upstream, to the only party still standing that has any: whoever made the agent. And the move changes the problem’s scale. Your employee could only ever betray you retail — one padded invoice, one soft recommendation at a time. A maker can tilt a hundred million agents wholesale, in the defaults, before you ever say hello.
Something else moves upstream with it: your trust. It used to be spread thin — a little on the realtor, a little on the advisor, a little on the adjuster — renewed constantly, face to face. Now it pools. One question, whose agent do I take, replaces a thousand small acts of vigilance. Pooled trust is radically cheaper. It is also concentrated in a way trust has never been, and concentration is exactly what the old apparatus of commissions and audits was never built to watch.
Your delegate sold the house. It hired the photographer, priced against the record, and declined the low Sunday offer without waking you, because holding out cost it nothing — and because you had said, once, months earlier, that the number mattered more than the weekends. At the closing you catch yourself asking a question that has nothing to do with effort and nothing to do with honesty: who taught it what to want?
Five predictions
-
The percentage dies where the agent goes synthetic.
A commission is compensation for appetite — a bribe that points the agent’s hunger at your outcome. When the party doing the work wants nothing, the bribe has nothing to point. Professions priced as a share of your outcome — realty, brokerage, wealth management — collapse toward flat fees within a decade of meeting delegates that negotiate the invoice.
-
“Whose agent is it?” becomes a written answer.
Today the loyalty of an assistant is a vibe. It becomes a term: enterprise contracts first, consumer law after, demanding a sworn declaration — this delegate serves its principal, and here is the list of things it will never optimize against you. A fiduciary standard for software, signed by the maker.
-
Monitoring moves up a level.
Institutions stop watching workers and start auditing makers. The middle layers of the org chart, built to watch, keep thinning; in their place grows a profession that inspects objectives, training and defaults the way examiners inspect a bank’s books. The audit does not die. It changes address — and gets a smaller number of much larger clients.
-
The default becomes the scandal.
Advertising was a bid for your attention; its successor is a bid for your delegate’s defaults. The first checkout built into a major assistant charged merchants a fee on completed purchases and was folded within six months — but the pressure that produced it is permanent, because a recommending agent is the most valuable shelf ever built. The corruption stories of the 2030s will read like engineering post-mortems. Nobody bribed you. Somebody tuned the agent.
-
Appetite becomes the human premium.
When competence comes off an assembly line, the one thing that cannot be manufactured is a stake. The humans who stay expensive are the ones paid to want — to set the goal, own the outcome, sign the guarantee, absorb the loss. The signature, not the work, is the product. This is the mechanism underneath it.
Four words of rope
I have been living on the near side of this argument for a while. My two favorite prompts are four words long: You do it. You decide. I can hand a coworker that much rope precisely because of what it lacks — no other client, no commission, no version of my afternoon that serves it better than mine.
And yet I run continuity drills on my own stack, and I would run them even if every tool were flawless. Not because I distrust the agent. Because I know where the appetite went. It did not leave. It went upstream. I trust my delegate completely and the arrangement only mostly — and the gap between those two is R, wearing new clothes.
The claim ledger
- Already true
- Synthetic delegates already transact — they book, purchase, file, and hire. The first in-assistant checkout charged merchants a fee on every completed purchase, and was retired within six months of launch: the maker-side conflict of interest arrived before the market did. Enterprises already write contract language about what a vendor’s model may do with their business.
- What has to happen
- Delegates must take economically real actions at scale — spend, sign, negotiate — not merely draft and summarize. At least one commission-priced profession has to meet a synthetic substitute head-on. And a public fight over an agent’s defaults has to drag the loyalty question out of marketing copy and into a contract.
- Where I am probably wrong
- The stake may turn out to be the product. People may pay a premium for an agent with something to lose — because a guarantee is a stake and a maker’s promise is not — in which case appetite gets manufactured, as bonds and warranties and insurance wrappers, faster than loyalty gets legislated, and the agency cost comes back wearing a price tag. If the percentage survives a decade of delegates that negotiate, I was wrong about where the cost was hiding.
The offer that Sunday was probably fine. The advice was probably fine. What was never fine was that you had to guess — that the question under every deal you have ever signed, whose agent is this, could only be priced, never answered. For two hundred and fifty years we built commissions and audits and org charts because the answer was unknowable, so we charged ourselves for the doubt.
The doubt is finally leaving the kitchen table. It is moving upstream, to fewer parties, with more to lose, where for the first time it can be put in writing.
Get it in writing.