Two Ways a Delegation Dies — How Courts Spent a Century Dismantling Instructions That Said "Never Change This"

Section 7 of the first piece originally carried a long stretch of family-office history. Compressed, that section came down to two surnames, and what I cut was the part of this series I am most confident about: it is not my speculation, it is a hundred years of court record.

So here it is on its own.

The first four pieces were all about things that have not happened yet: a paradigm, an implementation, a defect in the harness, a page of vocabulary. This one is about things that already happened. Handing the decisions of your life wholesale to a set of agents is an experiment humanity has run for at least a hundred and fifty years, the sample is trusts and family offices, and the failures have files, figures and judgments.

Having read them, my conclusion is that the failures are not randomly distributed. They land, over and over, in the same two places.

1. The experiment already has a sample

#

The first piece said that once the parenting agent goes live, you are promoted to principal, and that the principal’s one core competence is judging whether the agent is working for you.

I wrote that as if it were a new problem. It is not. The whole discipline of trust law is about exactly this, and has been for a long time.

Robert Sitkoff wrote An Agency Costs Theory of Trust Law in 2004, which takes the trust apart as a standard agency problem: managerial authority sits with the trustee, the residual claim sits with the beneficiaries, the two are separated, agency costs follow — plus a temporal layer, whether the trustee stays loyal to what the settlor originally meant.

Two sentences in it made me sit up:

“Beneficiaries are often unsuited to monitor the trustee, perhaps because they are unborn, incapacitated, or simply irresponsible.”

“Beneficiaries are not normally thought to give ex ante consent, and typically they are in no position to bargain.”

His conclusion is that fiduciary duties exist because they are substitutes for monitoring by the directly interested parties — you need duties precisely when the person with the direct interest cannot do the monitoring.

That is about trusts. But swap trustee for agent and beneficiary for user and not one word needs changing. And it is twenty years early.

2. The first death: the instrument locks, and the world moves

#

Pulitzer: he knew how to grant a power of sale, and deliberately withheld it

#

Joseph Pulitzer’s prohibition is not in the 1904 will. It is in the first codicil, dated 23 March 1909. That matters, because of how the codicil is written.

In the same clause he authorised his trustees to sell the stock of the Pulitzer Publishing Company of St. Louis — “at any time, and from time to time… at public or private sale, at such prices and on such terms as they may think best” — and said expressly that the power “is not to be construed as in any respect mandatory, but purely discretionary”. And then, immediately:

“This power of sale, however, is limited to the said stock of the Pulitzer Publishing Company of St. Louis, and shall not be taken to authorize or empower the sale or disposition under any circumstances whatever, by the Trustees of any stock of the Press Publishing Company, publisher of ‘The World’ newspaper. I particularly enjoin upon my sons and my descendants the duty of preserving, perfecting and perpetuating ‘The World’ newspaper (to the maintenance and upbuilding of which I have sacrificed my health and strength) in the same spirit in which I have striven to create and conduct it as a public institution, from motives higher than mere gain.”

One document, holding “you may sell” and “under no circumstances whatever may you sell” side by side. He did not forget. He demonstrated a delegation and then withdrew it in the next sentence.

Twenty years later the trustees were his three sons (Ralph, Herbert, Joseph Jr.), and there were fifteen remaindermen, fourteen of them minors.

Then the numbers went wrong. All of the following is from the opinion itself:

The judge noted one thing specifically: the decline in revenue was not caused by the 1929 crash, but “antedated it by at least two years”. This was not an accident; it was a structural decline, and the instrument forbade the response.

Nor did the trustees sit still. The opinion is explicit:

“The trustees have attempted to correct the deterioration which has occurred by employing specialists and experts in the advertising and circulation fields.”

They hired advertising and circulation experts. It did not work. Because the one thing they were forbidden to do — dispose of the asset — was the only option left.

Late on the night of 26 February 1931, Surrogate James A. Foley of New York County ruled. His reasoning ran like this:

“The dominant purpose of Mr. Pulitzer must have been the maintenance of a fair income for his children and the ultimate reception of the unimpaired corpus by the remaindermen… A man of his sagacity and business ability could not have intended that from mere vanity, the publication of the newspapers, with which his name and efforts had been associated, should be persisted in until the entire trust asset was destroyed or wrecked by bankruptcy or dissolution.”

“The law, in the case of necessity, reads into the will an implied power of sale.”

Notice what he actually did: he used his inference of the settlor’s real purpose (grandchildren receiving an unimpaired corpus) to override the settlor’s written command (under no circumstances whatever).

He knew how large a step that was, which is why he first cited an almost identical precedent: The New York Times, 1893. George Jones’s will forbade his executors to sell his 46 shares of the New York Times Association; the paper became unprofitable; on 8 August 1893 Justice Morgan J. O’Brien authorised the sale on the same reasoning. The World was the second great New York paper prised out of a dead hand by the same argument.

Three things the popular version usually gets wrong, and they happen to be the point.

First, Foley never once used the words “deviation” or “dead hand”¹. Those are labels later textbooks attached. What he used was an implied power of sale read in by necessity, plus the power of a court of equity in emergencies.

Second, the court did not approve the sale to Scripps-Howard. Foley twice stated he had no jurisdiction over it, on the ground that a surrogate’s court cannot reach the internal affairs of a corporation; he also wrote a hard line about how no “unwarranted use of judicial power” could “justify the conversion of the court into an auction room for bidders”, and said the selection of the purchaser, the price, the payment terms and the buyer’s credit rested entirely on the corporation’s officers and directors.

Put those two together and the picture changes. Foley lifted the prohibition — and in the same judgment pushed “to whom, and for how much” back onto the three sons, who happened also to be directors of the company, and happened to be interested parties in the transaction. The dead hand let go of the wheel, and the hands that took it had a conflict of interest.

Third, the price. At 1:30 a.m. on 27 February, Roy W. Howard took the bill of sale from the three brothers. The nominal figure was five million dollars, but the structure was: $500,000 in cash, $500,000 at ninety days, $2,000,000 in notes, and a further $2,000,000 contingent on the profits of the merged paper. Counsel for the employees said in open court that this was, in effect, “a definite offer of only three million dollars”.

The three papers employed about 3,000 people. On the evening of 25 February, 450 of them met at the Hotel Astor and formed the World Employes’ Co-operative Association, pledging over a million dollars to buy the papers. It failed. A week later the association disbanded; between 400 and 500 were taken on by the new World-Telegram, and the trade estimate was that about half of the former staff found positions.

One last thing, for anyone who thinks Foley rescued everybody: in 1984 the Florida Law Review called this “the leading case in this area” and then attacked it hard — saying that for a court to declare the testator’s main purpose to be income for his sons and an unimpaired corpus, in the face of language prohibiting sale “under any circumstances whatever”, “seems mere sophistry”.

I am keeping that, because it is honest: “overriding a mandate that has gone out of date” and “an agent redefining what the principal wants” are technically the same move. That is exactly what the guard rail in section 5 of the first piece was trying to block — and here we see that even when a court does it, it gets accused of making the story up.

Barnes: not one override, but sixty years in slow motion

#

Pulitzer was a single event. The Barnes Foundation demonstrates something closer to the agent case: loosenings accumulate, and each one uses the last as its justification.

Albert Barnes’s indenture was locked tighter than Pulitzer’s. A few of the provisions:

• ¶27 (as amended 1941): after his death, funds may only be invested in obligations of the United States, the several states, and municipal corporations — limited to what are “legal investments for saving banks under the laws of the State of New York”.
• ¶33: after his death, “at no time… shall there be held in any building or buildings any society functions commonly designated receptions, tea parties, dinners, banquets, dances, musicales or similar affairs” — with a citizen-suit clause paying the total legal expense of any Pennsylvania citizen who sued to enjoin a violation.
• ¶30 (as amended 1946): the gallery and arboretum open to the public on Saturdays only, 10 a.m. to 4 p.m., except July and August.
• And: “no picture belonging to the collection shall ever be loaned, sold, or otherwise disposed.”

Then, one step at a time:

Ott’s own summary of the 2004 holding:

“…the provision in Dr. Barnes’ indenture mandating that the gallery be maintained in Merion was not sacrosanct, and could yield under the ‘doctrine of deviation,’ provided we were convinced the move to Philadelphia represented the least drastic modification of the indenture that would accomplish the donor’s desired ends.”

“Least drastic modification” is a good standard. The problem is that it is satisfied every time — because the baseline for each step is the state left by the previous loosening.

And there is a lesson here for agent design that lives entirely in the chronology: the thing strangling the endowment was the ¶27 investment restriction, and it was lifted in 1995. The move was 2004. The lock that caused the financial crisis had already been opened, nine years before the crisis was used to justify the move.

Two more items, because they show that “a court ruled” is not the same as “a fact was established”:

• On cross-examination in 2004 it emerged that the Foundation’s actual operating deficit for 2003 was about $1.2 million, significantly lower than the figure projected by Deloitte in 2002 and relied on at the earlier hearings. That is not from the opposition; it is in the court’s own opinion.
• The sole appeal from the 2004 decree was quashed as untimely on 27 April 2005. So that decision was never reviewed on the merits by an appellate court. Later petitions in 2007 and 2011 were dismissed for lack of standing.

Buck and Hershey: two counterweights

#

To stop this becoming “the court will always come and save you”, here are two that run the other way.

The Buck Trust. Beryl Buck died in 1975; the Tenth Clause of her will required the residue to be used “in providing care for the needy in Marin County, California, and for other non-profit charitable, religious or educational purposes in that county”. She left 69,156 shares of Belridge Oil. In December 1979 Shell bought Belridge at $3,665 a share, $3.65 billion in total — and that holding became $253,456,740. A modest county-level charitable trust suddenly had to spend a quarter of a billion dollars inside one of the wealthiest counties in California.

The San Francisco Foundation petitioned on 30 January 1984 for cy-près², wanting to spend part of the income in the other four Bay Area counties. It did not argue that Marin’s needs were met, nor that compliance was impossible or illegal — only that compliance was “impracticable, inexpedient and inefficient”.

The outcome: on 28 July 1986 the petitioner withdrew its own petition and asked to resign. But the intervening objector-beneficiaries were not parties to that settlement, so the court had to decide anyway — and on 15 August 1986 held cy-près inapplicable and refused the modification. No appeal was taken.

The cost: over $12 million in court-approved fees, paid out of the trust. And the rigidity is still running: the foundation reported total assets of $1,133,444,734 for the year ending June 2025.

The Milton Hershey School Trust overturned a premise I started with. I had wanted it as an example of a will locking in a single stock. It is not: ¶5 of the 1909 deed empowers the trustee to invest in “any securities which the Trustee and the Managers together may consider safe”. The concentration came from the 1918 gift, from subsequent practice, and from politics — not from a written command.

But the result was the same. By 2002 the trust held about 77% of Hershey’s voting power (today roughly 80% of the votes on under 30% of the equity). When the trustees moved to sell, Pennsylvania Attorney General D. Michael Fisher — at that moment his party’s nominee for governor — petitioned and obtained an injunction. On 17 September Wrigley bid $89 a share, about $12.5 billion, a premium of roughly 42%; that evening the board voted 10 to 7 to reject both bids and end the process. The next day the Commonwealth Court backed the Attorney General, holding he could inquire whether a trustee’s exercise of a power — “even if authorized under the trust agreement” — was “inimical to the public interest”.

Klick and Sitkoff ran the event study in 2008: a +25% abnormal return on announcement, −12% on cancellation. Blocking the sale destroyed about $2.7 billion in shareholder wealth to preserve roughly $850 million in charitable-trust agency costs.

Together these two say: a lock does not have to come from the settlor’s written words. It can come from a structure in which nobody has the authority to unlock it. Hershey had no such command, but it had voting power badly decoupled from economic interest, plus a single supervisor facing an election. The result was the same as writing it down.

The law’s own position on this is quite clear

#

Section 412 of the American Uniform Trust Code addresses exactly this. The provision itself is technical, but its official Comment contains the cleanest statement of the principle that the dead hand has a boundary:

“An owner’s freedom to be capricious about the use of the owner’s own property ends when the property is impressed with a trust for the benefit of others. Thus, attempts to impose unreasonable restrictions on the use of trust property will fail.”

The Comment is careful to draw the boundary narrowly at the same time: the purpose of equitable deviation is “not to disregard the settlor’s intent but to modify inopportune details to effectuate better the settlor’s broader purposes”.

And the decisive one is in §105: a settlor may override a great many default rules, but may not override the court’s power to modify or terminate a trust under §§410–416. Which is to say: “this may never be changed” is a sentence the law does not let you write.

That clause should be copied straight into the design of an agent’s mandate. The rules you set when calm may be very hard to change, but you must leave one channel that they themselves cannot close. Pulitzer’s mistake was not that he spoke in absolutes. It was that he never wrote down what should happen after three straight years of losses — he removed the judgment and left no trigger for review.

3. The second death: you cannot read your own structure

#

The first death is at least visible: the newspaper is losing money, the paintings are on the wall. The second is harder, because its symptom is that you cannot see the symptoms.

Pritzker: the information came only through litigation

#

The Pritzker family held Hyatt and other assets inside hundreds of domestic and offshore trusts — by some accounts more than a thousand. The skeleton is visible in Hyatt’s own public filings: one set of U.S.-situs trusts with Thomas J. Pritzker, Marshall E. Eisenberg and Karl J. Breyer as co-trustees; another set of non-U.S.-situs trusts with CIBC Trust Company (Bahamas) as trustee. In the FY2009 annual report the former held 46.7%, the latter 13.7% through an intermediate company.

The sequence runs like this, and it is not one transaction but at least three rounds:

• 1994: a change of trusteeship makes Robert Pritzker trustee for his children Liesel and Matthew. He moves the H Group Holdings stake (the Hyatt holding company) out of their trusts and into the Pritzker Foundation — a foundation he also controlled — which then redeems the shares back to H Group.
• 25 September 1996: ICA, the offshore bank, borrows $34 million from three of their T-551 trusts against a note paying 8%. Two days later, all the assets in Liesel’s and Matthew’s trusts — including that loan to ICA — are purchased by their cousins’ trusts for $150 million: $30 million in cash, the remainder in promissory notes paying 7.5%³.
• 2001: the assets are moved again, into new Texas trusts set up by a Houston attorney.

At that point the family said the two children’s trusts were worth $160 million each. The sentence Forbes wrote then is the important one: those sums “consist in large part of unsecured promissory notes”.

Which is to say: on paper you are wealthy, and what you hold are IOUs written by your own relatives.

And then the thing I find most telling, which is not in the news coverage but in a published appellate opinion.

In September 2002 the trustees petitioned a court to declare the 2001 family agreement to be in the best interests of minor and unborn beneficiaries. The petition stated its own motive: because the trustees’ determination “might not be conclusive if the minor and unborn beneficiaries could, years hence, raise their own claims against the trustees”.

The petition was filed under seal, and the agreement itself was never attached — only submitted for in camera review.

It took the Chicago Tribune intervening to get the file opened, and the appellate court held that the wholesale sealing was an abuse of discretion:

“The right to access… may not be evaded by the wholesale sealing of court files.”

In plain terms: the agents went to court to foreclose, in advance, the right of future beneficiaries to ask questions — and did it where those beneficiaries could not see.

As for Liesel herself. She filed in Cook County in November 2002; her brother Matthew joined in April 2003. The ruling of 5 March 2004 is often misread as a loss — the same ruling was headlined “Suit Gets Go-Ahead” by Forbes, “Judge dismisses Pritzker suit” by Crain’s, and “2 young siblings lose their inheritance lawsuit” by the AP. What actually happened: the judge granted 28 days’ leave to amend, sustained the core claim for breach of the duty of loyalty, and lifted the discovery stay.

They settled in January 2005 and jointly moved to dismiss. No court ever determined that anyone did anything wrong. Robert Pritzker denied the allegations throughout. Reported settlement figures vary widely; the defensible statement is that each of them received roughly $450–500 million in cash and trust interests.

Her explanation of why she sued comes from a Forbes interview in November 2003 — which, by Forbes’s own account, was the only interview she gave about the case⁴:

“I filed because I wanted to know what happened. It’s going to be tricky, and it will take a long time. But I just need to know what happened.”

In the same piece she also said: “This is not about cash.”

I want to be honest about this case: what she pleaded was that assets had been transferred away too cheaply, not that she could not understand them. And because it settled, no court ever characterised the structure at all.

But one thing is certain: the only route by which she obtained that information was filing a lawsuit, surviving a motion to dismiss, and waiting for discovery to be unfrozen. When a beneficiary has to invoke the judicial system to ask what happened to her own assets, the effective level of auditability in all the years before that is zero.

McNeil: he did not know he was a beneficiary

#

If Pritzker seems extreme, McNeil v. McNeil (Delaware Supreme Court, 2002) is a much cleaner sample.

A man was a current beneficiary of a trust, and the trustees did not tell him. The court:

“a trustee must communicate essential facts, such as the existence of the basic terms of the trust. That a person is a current beneficiary of a trust is indeed an essential fact.”

The duty applies without any request being made; his “repeated attempts to get information” should themselves have alerted the trustees that he did not know he was a current beneficiary; and the corporate trustees “should have known better”. The lower court’s description of how the trust had been run: on “autopilot”.

The crucial technical detail: the exculpation clause covered ordinary negligence — but not the duty to inform.

Hershey, part two: the beneficiaries had no standing to sue

#

Back to Hershey, but not for the money. In 2006 the Pennsylvania Supreme Court held that the alumni association lacked standing to sue the trust. “Private parties generally lack standing to enforce charitable trusts”, and the trust “did not contemplate the Association, or anyone else, to be a ‘shadow board’”.

So the beneficiaries of that school have no legal route whatsoever to demand an accounting from an endowment now worth about $17 billion. The only supervisor is the state Attorney General. And we have just seen that in 2002 that Attorney General was running for governor.

Handing auditability to a single supervisor you cannot remove, who has motives of his own, is the same as having no auditability.

The strongest evidence is the law’s own retreat

#

This was the most surprising thing I found.

Section 813 of the Uniform Trust Code sets out the trustee’s duty to inform and report. It looks thorough: keep beneficiaries “reasonably informed about the administration of the trust and of the material facts necessary for them to protect their interests”; supply the full trust instrument on request; give notice within 60 days of accepting a trusteeship; and at least annually, report the trust property, liabilities, receipts, disbursements, the source and amount of the trustee’s compensation, and a list of assets with market values.

The official Comment then reveals the actual water level:

• The duty runs only to “qualified beneficiaries”; remote remaindermen need not be given the information “unless they have filed a specific request”.
• “Ordinarily, the trustee is not under a duty to furnish information to a beneficiary in the absence of a specific request… the duty articulated in subsection (a) is ordinarily satisfied by providing the beneficiary with a copy of the annual report.”
• The drafters deliberately used “report” rather than “accounting”, “in order to negate any inference that the report must be prepared in any particular format or with a high degree of formality” — and the Comment goes so far as to say the duty “might even be satisfied by providing the beneficiaries with copies of the trust’s income tax returns and monthly brokerage account statements”.
• Can a trustee assert attorney-client privilege against a beneficiary? The drafters left it blank, to be resolved by the courts later.

And then the decisive fact. The 2000 version made two duties non-waivable by the settlor: notifying beneficiaries aged 25 or over of the trust’s existence, the trustee’s identity and their right to reports; and responding to a beneficiary’s request.

In 2004 the Uniform Law Commission put both of those in brackets, making them optional.

The reason was that they would not achieve uniformity — which is to say the states would not take them. Later study found that among the states adopting the code, not one enacted those disclosure provisions verbatim.

Ohio’s statute is the clearest specimen: the settlor may designate one or more “beneficiary surrogates”, and all notices, information and reports go to the surrogate “in lieu of providing them to the current beneficiaries”.

I set this out in full because it is so close to where we are standing: the designers of a system wrote “the principal must be told” into its mandatory provisions, downgraded it to optional four years later under practical pressure, and the market then declined to adopt it at all.

If auditability for agents goes the same way, it will look exactly like this: a beautifully written spec, then an option, then nobody turning it on.

The thing I could not find is itself the finding

#

What I originally went looking for was rigorous empirical evidence about whether beneficiaries actually read, or can parse, their own trust accountings.

There is none. What exists is practitioner assertion, wealth-transfer surveys measuring something else, and some qualitative interviews.

That is worth writing down on its own: the entire enforcement theory of trust law rests on the assumption that beneficiaries read the report, understand it, and act on it. That assumption appears never to have been tested.

And we are about to install the same assumption inside a system that speaks a few hundred sentences a day on your behalf.

4. Rockefeller: they delegated the work, not the supervision

#

All bad news so far, so here is one that has lasted.

The Rockefeller family office was founded in 1882, the first single-family office in the United States; from 1933 it occupied the 56th floor of the RCA Building, the famous “Room 5600”, and at its peak employed as many as 200 people. The family is now in its seventh generation, with more than 250 direct descendants.

What is worth learning is not “they hired excellent professionals so the descendants did not have to pay attention”. It is the opposite — they spent a great deal of effort keeping themselves able to pay attention:

• A family forum. David Rockefeller Jr.: “We meet as a family twice a year, often more than 100 of us in a same room… When you are 21, you get invited to those meetings.”
• Board seats. The Rockefeller Brothers Fund has 18 trustees, about half of them family members, seven from the fifth generation. And — this is the load-bearing detail — the role of chairperson is reserved for a family member, although this is an informal policy not specified in the by-laws.
• Succession training as its own institution. In 1967 the five brothers set up the Rockefeller Family Fund, whose stated purpose was “as a means of training their children (the ‘cousins’) and the family’s fifth generation in trusteeship and foundation philanthropy”; the RBF simultaneously created two non-voting “visitor” seats for the next generation to rotate through. The first cousin chaired in 1987; the first fifth-generation chair took over in 2013.

I have to qualify this honestly. When Rockefeller Capital Management launched in 2018, control of the operating business went to Viking Global, with the family retaining a minority stake and board seats. Chief executive Greg Fleming: “We’ve had two members on our board from the beginning to this day: David Rockefeller Jr. and Peter O’Neill.”

So the accurate statement is: what they supervise is the institutions and the governance structure, not the day-to-day operations of an asset manager. But that is exactly the point — they handed over the operating, and kept the standing and the capacity to supervise it, and built a dedicated institution to train the next generation for the job.

Of everything I found, this is the only case where “maintaining the principal’s ability to supervise” is treated as a line item worth investing in.

5. Translating this into an agent’s mandate

#

Compressing the files above into things you could write into a mandate.json:

One. A hard rule must ship with the conditions that force it to be reviewed. Pulitzer’s error was not speaking in absolutes; it was writing the prohibition and no trigger. UTC §105 is worth copying: a settlor may disclaim nearly every default rule, but may not disclaim the channel that overrides him. For an agent: the rules you set when calm should be very hard to change, but there must be one review path they cannot close, and its trigger conditions must be written while you are still calm — not negotiated in the moment the impulse arrives.

Two. Auditability has to specify a format, or it is a waiver. The lesson of UTC §813 is blunt: the drafters chose “report” over “accounting”, and the Comment then concedes that tax returns plus brokerage statements might satisfy it. Freedom of format is exemption in substance. So an agent’s audit output cannot just be “there are logs”; it has to answer one specific question: why this number, on this line? That is what the never-leaving private rationale field in the implementation piece was doing, and I now think it is not an optional extra.

Three. Accessible is not the same as intelligible. This is the real lesson of the Pritzker section. She had $160 million on paper and the data was not entirely absent — the problem is that the $160 million was “in large part unsecured promissory notes”. The information was there; the understanding was not. An agent’s window that shows you only a final balance is structurally the same object as those notes.

Four. Do not give the only supervisory power to someone you cannot fire. The Hershey beneficiaries had no standing, and the sole supervisor had an election to win. For an agent: if the audit record lives only at the vendor and you hold no local copy, your supervision exists at someone else’s pleasure.

Five. Treat “staying able to understand this” as a thing you invest in. That is the Rockefeller clause. Section 6 of the first piece asked which layer you want to occupy yourself; here is a more basic one: which comprehension do you intend to keep, even where outsourcing it would be more efficient?

6. Where I might be wrong

#

The analogy has one crucial asymmetry. The beneficiary of a trust is usually not the settlor — Pulitzer’s grandchildren, Barnes’s posthumous public, Liesel. The principal of a parenting agent is the user. That makes revocability enormously easier: you can fire your agent at any time, and a trust beneficiary cannot. So the first death threatens agents less than it threatens trusts.

But that in turn makes the second death purer. A trust at least has an external supervisor — a court, an attorney general. Between you and your agent there is none. Nobody else is going to read that audit file for you.

Everything I cite is American law. Agency and trust law elsewhere differs, and the legal conclusions here do not transfer directly. I am using these as engineering case studies, not legal advice.

The Barnes “financial collapse” finding is contested on the record, which I flagged in the text. Buck is an unappealed trial-court decision: persuasive, not binding.

The biggest hole is still where it was. This piece is about the vertical relationship between a principal and their own agent. The horizontal problem from section 4 of the first piece — two well-behaved agents negotiating, with the surplus flowing to whichever is more aggressive — gets no help here at all.

The first piece said that once the parenting agent goes live, you are promoted to principal. This one went and looked up the service record for that position.

The conclusion is less comfortable than I expected: across a hundred and fifty years, handing the work over was never the cause of failure. The cause was that after handing over the work, the principal also dispensed with the supervision — and at some point discovered they no longer had the capacity to take it back.

Foley’s sentence was really addressed to every principal, not only to Pulitzer: a person of judgement could not have intended their own written rule to be executed until the thing they meant to protect was destroyed.

The difficulty is that the only person who can see it has reached that point is the one still looking.

Every term this series coins or borrows is collected in the glossary. Earlier pieces: the paradigm, the implementation, principal collapse.