Summary

Exceptional investment partnerships begin by collapsing the distance between principal and agent. Stuart Miller first entrusted Graham Duncan with $50 million, expanded the mandate after a year, and kept rooting for him even when Lennar fell near $3 and East Rock lost roughly 12% during the financial crisis. That “we’re grown-ups and we’re taking risk” posture gave Duncan permission to invest for outcomes rather than optics: “This is a game. We’re playing a game. It just got interesting.”
Duncan’s preferred family-office CIO is commercial, quiet-egoed, and conservative enough to survive. Commercial means possessing “the ability and the intent to create more value than you capture,” playing a repeat game, and preferring to make money rather than prove oneself right. Quiet ego permits returns to come through other people; conservatism supplies the visceral refusal to sell puts, court ruin, or otherwise leave the game.
Manager seeding only works when capital follows an initiative the manager already owns. Duncan would back a manager only if he would invest without seed economics—at a smaller size and without shaping the firm—and warns that a dedicated pool “burning a hole in my pocket” corrupts selection. The deepest diligence question is “Who is source, and are they owning it?” because breaking the shell for an aspiring founder can prevent the strength that comes from taking the first risk.
The best talent is often mispriced precisely because its current institution has reasons not to validate it. Duncan looks for young managers with limited public evidence, former bosses damning departing stars with faint praise, and the spread between what an internal platform lets someone run and what outside investors would initially entrust to them. The candidate is often asking one market question beneath the career discussion: “What’s my price?”
Compulsion is a more durable career compass than prestige, job description, or conventional productivity. Duncan can browse LinkedIn for hours, studying the relationship between a self-selected photograph, a résumé, and “what life has done to them”; Patrick reframes that apparent laziness as prolific pattern collection followed by a few leveraged decisions. The practical questions are “What are you compulsive about?” and whether a moment of ignition ever made you think, “I want to be that.”
Launching a firm requires holding ambiguity while protecting the psychological conditions for future decisions. A founder leaves an inherited reality for a world where investors, team, and office may each carry only a 33% probability, then must make the enterprise feel real enough that belief helps create it. Duncan therefore treats “the climate in your skull” as an asset: an anxious CFO, misaligned partner, or fast-twitch LP can transmit fear exactly when judgment matters most.
Deep referencing is both a selection tool and an operating manual for the relationship that follows. Duncan estimates its signal at roughly 75%, provided it reconstructs behavior across people, contexts, and time rather than collecting three approved calls. The investigator must distinguish the conscious “rider” from the behavioral “elephant,” notice their own reflection in the interview, and understand the institutional “water” that may have made prior success possible.

Deep dive

1. Trust turns an allocator from agent into owner

Duncan’s diagnosis of weak family-office structures is a principal-agent paradox: “If you’re the principal and you treat the agent like an agent, then they become an agent.” Commercial people want more control than most principals grant, so bringing an A player inside a tightly controlled boundary often suppresses the very behavior the principal hoped to acquire.
Stuart Miller instead let Duncan establish an independent company, initially allocating $50 million and allowing a year for trust to develop. That path may have prevented the relationship from becoming “run my family office, but it’s my thing,” creating an atmosphere of “let’s make money together” rather than one person managing another’s expectations.
Compatibility still mattered. Duncan argues that principal and allocator need sufficiently overlapping “maps of reality”—a mathematical principal may need a similarly left-hemisphere CIO—because every strategy eventually suffers adversity. One family office reportedly cycled through roughly 15 CIOs after repeatedly mishandling that moment of disagreement.
The shared long horizon was reinforced by Miller’s defining leadership quality: people around him felt he was rooting for them. Duncan connects this to Randall Stutman’s work on exceptional leaders and remembers the unmistakable message: “I’m rooting for you first and foremost. Let’s do this.”

2. The financial crisis revealed the value of a light grip

Miller gave East Rock the rest of his capital outside Lennar in early 2007; Duncan recruited Adam Shapiro from Goldman Sachs, leaving the young partnership barely a year before the financial crisis. By 2008, Lennar had fallen near $3 and risked bankruptcy while Miller’s money was with the young partnership.
East Rock was down about 12%. Duncan and Shapiro had anticipated parts of the crisis and expected to be flat or positive, so they became visibly dejected. Miller, simultaneously raising money near the bottom, challenged them: “You guys, what the hell do you think? I knew we were at risk of losing money. You don’t see me moping around.”
Duncan remembers Miller becoming physically jaunty, evoking “a SEAL team captain” rather than defending an identity as a billionaire. His message was: “This is a game. We’re playing a game. It just got interesting.” That lightness acknowledged genuine danger without turning losses into shame, leaving the managers better able to take necessary risk.
The rowing metaphor is the “right grip”: an oar must be held firmly but loosely enough to release if its blade catches underwater. Ideology tightens the forearm and makes a belief non-negotiable; humor loosens it. Miller joking about dragging bags of gold through the street conveyed, “We’re gonna do the best we can here, and who the hell knows what’s gonna happen.”

3. East Rock organized around comparative advantage, not a fixed product

Patrick characterizes East Rock as unusually organic: multiple strategies, manager structures, seeds, LP positions, and direct investments all emerged from the simple aim of finding exceptional investors and making good investments with them. Duncan agrees that the platform was downstream of a plainer mandate: “Make money and not lose too much.”
Duncan did not believe he possessed comparative advantage in stock-picking or most individual investment styles. His advantage was identifying people, understanding their circumstances, and configuring capital around them. With Shapiro running the platform, East Rock can morph around a different comparative advantage rather than preserve Duncan’s methods as doctrine.
That flexibility still required earned trust. Miller provided “a playing field with no constraints, but you had to earn the right to go anywhere.” Duncan’s hypothetical redesign would again fit the current opportunity set to a principal’s risk tolerance—even seeding someone with $1 billion if confidence and worldview alignment genuinely justified it.
The ideal removes the allocator’s need to demonstrate visible activity or authorship: every dollar should reflect what the agent would do if it were personally theirs. Duncan admits he has “zero—like negative” desire to work for work’s sake; the useful consequence is an aversion to activity undertaken merely to prove that an allocator added value.

4. Commerciality means creating more value than one captures

Duncan defines “commercial” as “the ability and the intent to create more value than you capture.” Its abundant form signals a repeated game: the actor does not seize every available penny because proportionality, reputation, and the expectation of meeting again matter. Goldman’s “long-term greedy” captures the orientation.
Another component is preferring money to vindication. Some investors primarily seek “the satisfaction of being right,” which can work until identity and portfolio become fused to a position. Commercial actors update because the goal is making money, not preserving an intellectual self-image after the evidence changes.
Duncan’s taste evolved accordingly. He once favored Tina Fey’s “Harvard nerds”—structured, stable, and meticulously planned—before learning to value more “Chicago improv,” whose practitioners will alter anything for the result. The optimal manager combines rigor with what Charlie Munger called “the knack”: plasticity without becoming merely fly-by-night.
A former Milken trader once put a hand on Duncan’s arm and said, “Money’s like water. All you have to do is learn to turn on the faucet.” Beneath the theatrical delivery was a serious principle: ask, in Joe Hudson’s phrase, “Where does the water want to flow downhill?” Pragmatism works with reality instead of forcing a cherished setup upon it.

5. Dan Sundheim exemplified mastery with a quiet ego

Duncan tracked Dan Sundheim partly through former colleagues at Viking, who spoke of him as professionals describing a master. One recalled Sundheim planning how to enter and exit a stock six months before buying it, combining fundamental analysis with an unusually developed understanding of liquidity and market structure.
Hedge-fund founders continually “price” rising investors through compensation and carry. Duncan watched for years in which the founder failed to “hit the bid,” creating an opening for the market to reprice an under-recognized manager. Sundheim’s decision to launch offered that kind of opportunity.
Duncan tried to earn a dialogue by helping Sundheim recruit analysts, eventually contributing perhaps four of the initial 12 hires—figures he explicitly says may be wrong. The process let him calibrate Sundheim’s judgment and observe how he absorbed difficult feedback with “no defensiveness whatsoever,” dropping an idea immediately when new information warranted it.
East Rock became a day-one investor. Duncan acknowledges the fund has experienced “its ups and downs,” but continues to believe in Sundheim as both a commercial actor and firm leader. The durable signal was not a frictionless record; it was the repeated choice of commercial reality over ego.

6. Seeding succeeds only when the investment works without the economics

Duncan’s seeding standard was counterfactual: “I would invest in this manager even if I didn’t have seed economics.” Without revenue share or ownership, he might allocate less, accept less risk, and avoid reputational responsibility—but the underlying investment still had to clear the bar.
Terms then price what the market might not otherwise supply: duration, lockups, fees, or other features of the arrangement. Seed economics are not a reason to relax manager selection. Patrick noted that in East Rock’s eight seed deals, basically all or most worked out very well; Duncan’s stated discipline was to view himself as backing people, not operating a seeding factory.
Patrick’s pushback is the model’s apparent seduction: LP capital funds the investment while the seeder receives GP economics, making it look like a “money machine.” Duncan’s answer is adverse selection plus deployment pressure. Give him $2 billion that must be seeded, he says, and he would not trust his own judgment because “I need to put assets out” ruins the dynamic.
Duncan prefers managers who would launch anyway. A prestigious investor once received an unsolicited offer of roughly $200 million; Duncan suspected the fund would never have been conceived without that call. Capital had manufactured the aspiration, changing the founder’s relationship to the enterprise from “my thing” into something closer to a well-paid job.

7. Source dynamics explain why promising institutions fracture

Peter Koenig’s source framework begins with the person who took the first risk—even if that risk was merely calling a future co-founder. Duncan’s diagnostic is: “Who is the source, and are they owning it?” He found one quant founder behaving like an employee because the enterprise had effectively happened to him after a friend initiated it. Koenig’s broader claim is that organizational dysfunction can be traced to disagreements about who is source or to the source failing to own it.
Diana Chapman’s chick analogy supplies the mechanism: the chick must peck through its own shell to build the strength required outside it. “If you break the shell for them, they will die.” Interfering with the origin can distort an organization’s trajectory long after conventional diligence has forgotten who initiated what.
Co-founder resentment often reflects an unacknowledged source hierarchy. Duncan estimates it may remain contained 80% of the time, then surface under failure or extreme success. Succession is similarly difficult because heading off source is subtle; the successor must coherently own the reality rather than merely receive the title.
Duncan’s own first company belonged to Yale professor Richard Medley, even though 21-year-old Duncan managed roughly 30 employees. Conflict arose whenever Duncan wanted “the Graham Show” while actually producing “the Richard Medley Show.” At East Rock, he later believes he sometimes failed to own source fully before coherently passing it to Shapiro.

8. Talent mispricing often hides in institutional incentives

Duncan likes very young managers because limited information creates legitimate inefficiency, much as it can around an IPO. A second favored pattern is a former boss damning the departing star with faint praise: endorsing that person too strongly could encourage every other employee to leave, so the boss has structural reasons to obscure quality.
Duncan has five to 20 conversations at a time with people considering departure and finds their fear of the incumbent platform’s reaction both consistent and justified. His test is whether a prospective founder can move beyond avoidance—“My job sucks”—and articulate a proactive vision larger than manufacturing a new job or copying friends who became founders.
At the highest level, the conversation becomes market pricing. A senior hedge-fund investor may really be asking whether an independent launch raises $1 billion or $5 billion. Another investor might command only $500 million to $1 billion outside his institution while successfully running more than $12 billion inside it; that spread is the founder’s talent arbitrage.
Patrick compares this appetite for “hair” to Founders Fund: once a deal becomes universally sought, the mispricing disappears. Duncan’s refinement is that institutional tension can be signal, but only when the investor understands why the former employer might behave strategically—not whenever conflict makes a candidate sound exciting.

9. A family-office CIO must fit the principal’s taste and tolerate invisibility

Duncan doubts a YC-style accelerator can reliably produce family-office CIOs. Unlike startup investing, where raw young talent can discover a new market, broad capital allocation generally rewards grizzled practitioners who have taken risk with their own and others’ money and possess the EQ to express bets through other people.
The first criterion is good taste in people “from the perspective of the principal,” not some universal ranking. Principal and CIO must admire overlapping sets of people or trust will fail under pressure. Duncan points to the Collison brothers’ distinctive circle of “alpha nerds,” public-policy thinkers, and figures such as Tyler Cowen as an aesthetic that attracts more of itself.
The second criterion is a quiet ego. Many allocators secretly wish they were the underlying GP and struggle to own an analyst’s or manager’s best idea without making it theirs. The CIO must care about net-of-fee results, understand personal comparative advantage, and appreciate another investor without requiring authorship, identity, or power from the position.
The third criterion is conservatism “in their bones”: experience losing money and an aversion to “selling puts” or taking any risk capable of ending the game. The principal wants the agent to treat the capital as personal wealth, which is likelier when the CIO already has enough money to have learned to relate to it that way—or possesses the temperament—to understand permanent loss viscerally.

10. Talent density becomes a self-reinforcing sourcing advantage

Duncan’s gatherings deliberately create talent density. His guest-list test is visceral: if forced to sit beside this person, would he be neutral, excited, or disappointed? He tries to make every answer “psyched,” producing the atmosphere of a great wedding where there is “not a bad seat in the house.”
A San Francisco investment-firm leader once wrote afterward, “Graham, you’ve restored my faith in humanity.” The business effect follows from the social standard: high-trust people discover one another, do business, and strengthen the gravitational field surrounding the convenor without every interaction needing to benefit Duncan directly.
This began as a manager-evaluation method. Put roughly 30 investors in a room pitching ideas, and Duncan found that “the room knew who the best people were.” His prior rankings sometimes held, but sustained interaction also surfaced unexpected pockets of quality; reading collective attention became another form of diligence.
Physical space supports the same flywheel. East Rock intentionally carried surplus offices, a café, a chef, and a communal table so up-and-comers could work there without bearing all their own overhead. Duncan calls providing space “such an easy arb”: subsidize the missing infrastructure, collect talented people, and increase the chance that useful collaborations emerge.

11. Paradigm emerged from following credibility before the field looked credible

Charlie Songhurst touted Bitcoin before East Rock bought a small position at roughly $300 per coin. As it appreciated, Duncan attended a crypto conference to decide whether the asset was real; many early participants struck him as “not serious people,” a common pattern when new fields initially attract whoever is available and willing to switch.
In a room of roughly 300, Duncan first followed a woman then leading crypto work at Facebook because credible people clustered around her. When Sequoia’s Matt Wong appeared, Duncan followed him instead. Wong and Fred Ehrsam were using the trip partly to assess a partnership, and Duncan’s airport drive with them became an inadvertent founder-diligence session.
Duncan experienced each as a field-agnostic commercial actor and helped interview CFO candidates. He strongly endorsed one woman while warning that her Enneagram-one vigilance might make Fred feel she thought he was “getting away with something.” If he resisted taking that personally, Duncan predicted, she would bring institutional credibility and become exceptional.
She ultimately became CFO and a third partner, helping build the firm. Duncan says he thinks Paradigm manages roughly $10 billion today and credits the team with committing real capital at multiple crypto bottoms. The example captures his model: identify the people, help construct the relationship, and share enough of the “before and after reality” to judge how the institution actually compounds.

12. Pricing talent means calibrating autonomy as carefully as money

Duncan treats economics, capital duration, transparency, control, and decision rights as components of one talent price. Elon Musk’s xAI fundraise is presumably “egregious” to investors—little control or transparency—because the market prices his credibility. A less-proven manager receives tighter constraints because the evidence supports a different degree of autonomy.
His fairness test is empathetic: if he had this track record, life stage, and network, would the agreement feel fair or exploitative? If it felt exploitative, he reduced his claim. Formal rights could still remain extensive, but tone and a “very light touch” made the practical field feel more open.
Partnership economics require equally careful calibration. A colleague who believes they deserve 40% after receiving 10% carries resentment “in the water”; one who actually wants 50% may be better supported for several years before launching independently. Failure experience can improve self-pricing because it teaches how difficult source-level responsibility is.
Patrick’s formulation is the key stress test: could the partner become resentful after success and decide the achievement was more theirs than the economics acknowledged? Duncan sees dynamic equity and carry systems as a fertile way to accommodate changing contribution, while noting that most such systems benefit from one person who ultimately makes the call. As the quoted formulation puts it, “A partner is somebody who shares the same level of risk that you do.”

13. Founders must assert reality without poisoning their own judgment

Every institution is an idiosyncratic reality shaped by its source and its leader’s definition of reality. Duncan extends Gell-Mann amnesia to organizations: people recognize the weirdness of the container they are leaving, then assume the next one will be “normal.” It will not; age and long residence in one system can make crossing into another leader’s worldview increasingly difficult.
Launching means holding numerous 33% probabilities simultaneously: investors, team, and office may or may not materialize. On Monday morning there may be only a seat in Starbucks and “no order.” The founder must perform the creative act of defining reality despite vertigo and the fear that the entire venture is fake.
The paradox is that founders “have to pretend it’s more certain than it is because your pretending makes it so.” Henry Schuck captured this during a major ZoomInfo acquisition as “I’m playing pretend business.” Hedge-fund managers may struggle especially because their default skepticism can turn upon their own enterprise and dissolve necessary conviction.
Protecting “the climate in your skull” therefore becomes risk management: “The main asset is your future decisions.” An anxious, seasoned CFO can transmit fear when a younger PM is down 10%; an LP accustomed to quarterly liquidity can pull a slower strategy toward a faster-twitch cadence. Team and capital must supply enough skepticism to prevent ruin without contaminating confidence.

14. Duncan cast a restaurant by solving the whole human setup

Danny Meyer’s observation that some sites are unattractive even when free helped Duncan recognize a rare Montecito location whose “car kind of wants to go there.” He wanted a neighborhood institution with Chez Panisse-like mind share—a culinary place capable of disproportionately strengthening a community’s sense of itself.
After roughly a year securing a long sublease, Duncan interviewed many chefs and rejected prestigious licensing arrangements. A famous chef’s tenth outlet lacked the presence he wanted; the person cooking needed either a first breakout opportunity or a life reason to inhabit the restaurant rather than treat it as another branch.
Chef Joel Vieland stood out as knowledgeable without the post-COVID cynicism Duncan heard elsewhere. Vieland had previously acted as general manager, hated it, and therefore priced that role correctly. Duncan paired his skeptical, excellence-oriented Enneagram-six style with Jane, a locally raised general manager and Enneagram eight whose stability might complement it. “We’ll see whether it works.”
Housing completed the offer. Duncan renovated what had been Montecito’s cheapest house, three minutes from the restaurant and near a great public school, so Vieland could move with his child and belong to the neighborhood. Bryan Schreier and his wife became involved as hospitality-minded investors, leaving the project less like a restaurant hire than a carefully designed life platform.

15. Masters remain flexible; stewards care for the surrounding system

Running Sohn is another casting exercise: Duncan and collaborators selected 32 speakers by asking whether each was “a credible threat at saying something interesting” and whether seeing the name would prevent an attendee from taking a call. He preserved founder Doug Hirsch’s actionable ideas and no-panel format while budgeting selectively for broader voices and five-minute lightning talks.
Duncan estimates perhaps 10 to 30 investment masters exist. The transition from professional to master occurs when an investor becomes source of an idiosyncratic style and identifies principally as “a money maker,” not as a manager of one sector or method. Tepper, Druckenmiller, and Soros illustrate different forms of durable, adaptive investing; macro’s broad opportunity set can support that longevity.
Durable excellence combines love of the game, humility, and “bordering on paranoia” about missing the new regime. Markets provide brutal corrective feedback: harden ideologically or drink one’s own Kool-Aid, and capital disappears. The investor who complained that the SEC’s 2008 short-selling ban was unfair revealed attachment to the old rules rather than readiness to make money under new ones.
A handful progress toward stewardship by tending the wider machinery: Duncan cites John Arnold’s public-policy work, and says that, controversy aside and from afar, Bill Gates’s early-COVID posture and Mitt Romney’s current orientation seemed to fit that category. He also cites Druckenmiller warning about debt as a modern bond vigilante. AI leaves Duncan “scared and excited”; commercial, pragmatic, aggressive, humble actors should still thrive, though perhaps in forms no longer called hedge funds.

16. Referencing separates behavior from story—and context from portability

Duncan learned professional referencing through Ted Seides and the Yale investment-office lineage, which might track down college roommates rather than stop after three approved calls. He rereads “What’s Going On Here with This Human?” to recover the right mood: curiosity, enjoyment, and humility about how little any observer initially sees.
He estimates good references carry roughly 75% signal, not certainty. Their value extends beyond hiring: a pattern discovered beforehand often predicts how the person will later behave and supplies instructions for working together. Jonathan Haidt’s elephant-and-rider metaphor separates the articulate self-description from the less-conscious behavioral pattern visible across agents, settings, and time.
Duncan’s own “elephant” tolerates open loops and ambiguity longer than colleagues may enjoy. He can hold contradictory evidence without deciding, “keep eating the grass,” then move with surprising speed once the pattern resolves. The trait is useful in high-stakes selection but can make him frustrating to collaborators who prefer immediate closure.
Interviewers must also see their reflection in the window: a nervous interviewer may hold their breath, induce tension in the candidate, then diagnose the candidate as anxious. Finally, they must “see the water.” A commercial Goldman Sachs operator from the 1990s might appear sharp-elbowed elsewhere because success was partly native to that container’s rules, relationships, and norms.

17. Early contact with excellence created Duncan’s operating confidence

Rowing gave Duncan nine national championships and an embodied belief that he could outwork difficulty. A single-scull race covers 2,000 meters in roughly seven to eight minutes, producing sustained pain and “nowhere to hide.” That memory later supported integrity under pressure: even after launching East Rock just before 2008, he believed hard work remained available.
His club trained adult national-team athletes, used fancy boats, filmed every practice, and reviewed technique continuously. Duncan internalized the confidence that focused training produces improvement. His father called him highly coachable; Duncan remembers little resistance to changing technique because the objective was simply to win.
Yale supplied the intellectual equivalent. In a 60-person Directed Studies program, seven students came from St. Ann’s; two read Plato in Greek, and Duncan initially mistook one for a teaching assistant. He called home fearing he could not compete, then improved through writing one paper weekly and discovered that he could meet a previously invisible standard.
A late-night inference from professor Ian Shapiro’s paper led Duncan to offer research help before being asked. He built a large Lotus Notes database on workplace democracy and became Shapiro’s apprentice. The meta-lesson is “seek excellence early”—not merely watching it, but helping, training, and “touching” a standard that permanently enlarges one’s map of what is possible.

18. Wilderness, surrender, and rooted support complete the investment philosophy

Shapiro’s connection to Richard Medley led Duncan to a class that ignited an interest in markets and politics. Their firm became what Larry Summers called a “private-sector CIA.” Duncan discovered that raising a research subscription from $400 to $20,000 monthly generated meetings, then learned to identify people credibly connected to information streams during the 1997–98 Russian crisis.
After selling his stake, Duncan entered a wilderness of failed ventures and changing email addresses. The lesson was patience without passive waiting: make socialized concerns about status and relevance visible, then trust that compulsion can meet demand. Michael Singer’s frame captures it: “You are blind, and you have to learn how to be blind.”
Duncan’s ideal next setup remains investment-shaped: a principal with overlapping taste, mutual capital at risk, almost no principal-agent gap, and freedom to hunt unconstrained opportunities. He wants “license to hunt big game” beyond his personal chip stack, but at this stage is also willing to let the world disclose where his highest use lies.
That use is often holding complexity around a high-stakes human decision—business partner, investor, spouse, nanny—and surfacing unexpected positives or negatives. Miller’s original kindness supplied the model: taking professional risk on Duncan and continuously rooting for him. Duncan now tries to transmit that gift: “I felt that and so I know it, and then I can pass it along.”

R1, OpenAI’s o3, and the ARC-AGI Benchmark: Insights from Mike Knoop on the Gradient Dissent Podcast A Sentence-by-Sentence Guide to DeepSeek-R1, Kimi K1.5 & OpenAI o1