The Pure Software Autopsy
De Gaulle was wrong. The graveyards are full of replaceable men. The indispensable ones are the reason a few organizations outlast all others.
In 1990, Reed Hastings founded a firm called Pure Software. It grew the way fast software outfits grow: doubling revenue, doubling headcount, doubling again. Hastings was not careful about hiring. He needed bodies and he hired bodies.
“I wasn’t careful about it,” Hastings said later, “and I would say talent density declined. That company, we went public in ‘95, got acquired in ‘97. And when I analyzed, looking back, what happened, one of the major things was declining talent density. And then, with declining talent density, you need a bunch of rules to protect against the mistakes. And that only further drives out the high-caliber people.”[1]
The sequence deserves a slow walk-through because it is the most important causal chain in organizational life. First, the quality of the average employee drops by a small amount. The drop is not alarming. One mediocre hire among twenty good ones shifts the average by five percent. The mediocre hire makes mistakes. The mistakes are small but visible. A process is created to prevent the mistake from recurring. The process applies to everyone, including the fourteen who would never have made the mistake in the first place. The best operators, who came to the firm because it trusted them and gave them autonomy, now find themselves filling out forms and attending review meetings designed for someone else’s incompetence. They leave. Their departure drops the average further. More mistakes follow. More rules follow. More departures follow.
Physicists have a name for this. The Second Law of Thermodynamics holds that entropy, disorder, increases in any closed system without energy input. Leave a room unattended and it accumulates dust. Leave a garden unweeded and it returns to scrub. Leave a roster unmanaged and the average quality drifts, inevitably, toward mediocrity. The drift is not a failure of leadership. It is physics. The only force that reverses it is the continuous expenditure of energy: deliberate hiring, continuous evaluation, the willingness to fire. Pure Software was a closed system. Hastings stopped investing energy in the roster. Entropy did the rest. The firm went public in 1995, was acquired in 1997 at a fraction of what it might have been worth, and Hastings spent the next decade building Netflix as an explicit repudiation of everything that went wrong.
The lesson Hastings extracted from the wreckage concerned individuals, not process or technology or product. The single variable that mattered most was the quality of the average mind in the room. Get that variable right and everything else could be figured out. Get that variable wrong and no amount of strategy, capital, or consultants could compensate.
At Netflix, Hastings built the inverse of Pure Software. The core commitment was specific and uncomfortable: “We would say, we’re not going to guarantee you a lot, but we’ll guarantee that we’ll always surround you with great people and have you work on hard problems. That was our core. You may not be happy, the hours may be long. But the essence of what we can do at work is hard problems with great people.”[1]
Twenty percent of Netflix employees left in the first year. Hastings accepted this because the ones who left were the ones who valued security over challenge. Each departure raised the average. Each increase in average attracted more gifted applicants. The spiral compounded upward.
The Keeper Test formalized the spiral. Managers were asked a simple question: if this individual told you they were leaving for another outfit, would you fight to keep them? If the answer was no, the individual was given a generous severance package and wished well. Hastings designed a specific emotional quality into the language. He replaced the word “teamwork” with “open and respectful and dedicated to making others better.” The distinction was precise: teamwork could mean being nice. Being dedicated to making others better meant being honest, even when honesty was uncomfortable.
Hastings described the model as a sports team, not a family. Families do not cut members who underperform. Sports teams do. A family keeps the underperforming sibling at the dinner table out of obligation. A sports team replaces the underperforming player because the group’s purpose, winning, takes precedence over any individual’s comfort. The model sounds harsh. It is. But it produces something that no amount of team-building exercises or corporate values statements printed on break room walls can replicate: the knowledge that every individual in the room earned their spot. That knowledge is the foundation of trust. And trust, in organizations as in markets, is the asset that compounds fastest.
Dario Amodei, who founded Anthropic, articulated the density principle in its purest form: one hundred brilliant, aligned minds beat one thousand where only two hundred are brilliant and eight hundred are average. The mechanism is not additive. It is multiplicative. The two hundred brilliant individuals in the thousand-person operation spend half their time managing, compensating for, or working around the eight hundred average ones. The hundred brilliant minds in the hundred-person shop spend all their time on the work. The productivity gap between the two is not five to one. It is ten or twenty to one, because the absence of friction compounds across every interaction, every meeting, every decision.
If you have managed a group of any size, you can name your weakest member right now. You have known for months. The question the Keeper Test forces is not whether you can identify them. The question is what you will do about it, and how many of your best will leave before you act.
The Price of the Best
In 1913, Ford Motor Company was hiring at the rate of fifty-three thousand workers per year to maintain a constant workforce of fourteen thousand. The turnover was 370 percent annually. Every year, Ford trained and lost more than three complete workforces. The cost of training each new worker, the loss of productivity during the learning period, the mistakes made by untrained hands on an assembly line that demanded precision: enormous and largely invisible.
On January 5, 1914, Henry Ford announced that the minimum wage at Ford Motor Company would be five dollars per day, roughly double the prevailing rate. The financial press called it madness. The Wall Street Journal accused Ford of crimes against capitalism, which remains one of the most accidentally revealing headlines in American journalism. Competitors predicted bankruptcy.
The data told a different story. In 1915, after a full year of the five-dollar day, Ford’s hire rate dropped from 53,000 to 6,508. Turnover fell by nearly ninety percent. The cost of training new workers collapsed. Productivity increased because experienced hands stayed long enough to master their stations. Absenteeism, which had been running at ten percent, dropped to less than half a percent. Ford spent more per worker but less per unit of output. The five-dollar day was the most profitable cost increase in the history of American manufacturing.[2]
Ford framed the decision as self-interest, not charity. “We wanted to pay these wages so that the business would be on a lasting foundation. We were building for the future. A low wage business is always insecure.”[2] The logic was circular in the best sense: higher wages attracted better workers. Better workers produced better cars. Better cars generated more revenue. More revenue funded higher wages. Each element strengthened the others.
The deeper insight, which Ford articulated but which took decades to be rediscovered, was that wages function as a broadcast. The employer who pays above market sends three messages simultaneously. First, to current employees: we value you enough to pay more than we must. Second, to prospective hires: we attract individuals who are good enough to be worth this wage. Third, to competitors: recruiting our workers will cost you more than you expect. Evolutionary biologists call this the handicap principle: a broadcast is credible precisely because it is expensive to produce. The peacock’s tail works as a mating display because growing it is wasteful. Ford’s five-dollar day worked as a recruiting broadcast because paying it was reckless. The recklessness was the point.
In 2015, Mark Bertolini discovered that eighty-one percent of Aetna’s employees were women, twenty percent of employee families were on food stamps, and twenty-five percent of employees’ children were on Medicaid. The outfit was one of the largest health insurers in America. Its own employees could not afford health care. The irony was not subtle, but Aetna’s board managed to miss it for years, presumably by never walking through their own call centers.
Bertolini raised the minimum wage to sixteen dollars per hour. The cost was seventy-five million dollars in the first year. What happened was what Ford’s data had already proven a century earlier: the cost of higher wages was more than offset by reduced turnover. Recruiting costs fell. Training costs fell. Customer satisfaction improved because the operators answering the phones had been doing the job long enough to know the answers. The stock price doubled.[3] Ford’s data from 1913 to 1915 and Aetna’s data from 2015 to 2017 prove the same equation a hundred years apart. Paying more was cheaper than replacing more. In 2015, Dan Price at Gravity Payments pushed the same logic to its public extreme, setting a seventy-thousand-dollar minimum salary for every employee at his credit card processing firm. Revenue tripled over the following six years. The playbook still works. Most boards still refuse to run it.
Andrew Carnegie achieved labor peace through the opposite mechanism. Where Ford raised wages to attract and retain, Carnegie crushed wages to maximize margins and accepted the churn as a cost of doing business. The two approaches produced the same short-term result: stable production. They produced different long-term results. Ford built an enterprise that attracted loyal workers who identified with the brand. Carnegie built an enterprise that relied on replaceable labor and institutional systems that functioned regardless of who operated them.
The distinction maps onto a framework from evolutionary biology that clarifies the choice. Elephants, whales, humans invest heavily in few offspring and depend on each one surviving. Oysters, mosquitoes, dandelions produce massive quantities and depend on statistics: enough will survive. Carnegie ran a high-volume, low-investment workforce: survival through sheer throughput. Ford ran the opposite: low volume, high investment per head, survival through retention. The entire talent management debate, reduced to its biological substrate, is a question of which reproductive strategy fits your environment. If the work is modular and the workers are interchangeable, the Carnegie model works. If the work requires skill, judgment, and accumulated knowledge, the Ford model works. Most organizations operate as if they are Carnegie shops when they are actually Ford shops, paying commodity wages for non-commodity work and then wondering why their best leave.
Jorge Paulo Lemann built a third model, one that would horrify most HR departments. He called his ideal recruits PSDs: poor, smart, deep desire to get rich.
The system had four components. First, below-market base salaries and no perks. No fancy offices, no executive dining rooms, no company cars. The message was deliberate: if you wanted comfort, you applied elsewhere. Second, bonuses of four to five times base salary if targets were hit. Total compensation was well above market, but the guaranteed portion was below it. The system selected for operators who believed in their own ability to perform. Third, partners were not given equity. They had to buy it, with their own money, at market prices. The purchase bound the partner’s personal wealth to the firm’s performance in a way that stock grants, which cost the recipient nothing, could not replicate. Fourth, promotions went to whoever produced results, regardless of seniority, background, or connections.
The result was an operation populated entirely by individuals who had skin in the game, whose personal wealth rose and fell with the firm’s. The HR industry has spent fifty years trying to build assessment tools that identify high performers. Lemann skipped the assessment entirely. He built a compensation structure so unappealing to comfortable applicants that they self-selected out before the first interview. Most organizations spend millions on employer branding to attract the widest possible pool. Lemann spent nothing on employer branding and got the exact pool he wanted. The consulting firms that charge six figures for talent strategy would find this professionally embarrassing, which is probably why none of them recommend it.
3G Capital, the firm Lemann co-founded, acquired Anheuser-Busch for fifty-two billion dollars, merged it with InBev, and then acquired Heinz for twenty-three billion dollars. The operating improvements at each acquisition followed the same playbook: replace existing management with PSDs, eliminate perks, cut costs, tie compensation to performance, and let the incentive structure do the filtering. Within two years of each acquisition, operating margins improved by double digits.
John D. Rockefeller anticipated the same principle a century earlier. In the 1870s, Rockefeller proposed that none of the Standard Oil principals take a salary. Their entire compensation would come from dividends and share appreciation. A principal who wanted income had to grow the enterprise. A principal who wanted to retire had to sell shares whose value depended on the outfit’s continued success. There was no way to extract value from Standard Oil without first creating value in Standard Oil.
Ford paid more to keep. Lemann paid less upfront to filter. Rockefeller eliminated salary entirely to align. Three different mechanisms, three different eras, same underlying physics: compensation design determines who stays, who leaves, and what the remaining group optimizes for. The architect who designs the incentive structure is designing the organism. Everything else is commentary.
The Sapping
In the summer of 2002, Ari Emanuel began a campaign against the William Morris Agency that would take seven years to complete.
Emanuel chose patience over confrontation, and patience proved lethal. He poached agents, one at a time. He took clients, one at a time. He leaked negative press about William Morris’s management. He cultivated relationships with individual board members. He offered better deals to clients that William Morris considered locked in. No price wars. No lawsuits. No mergers. Just the slow, quiet erosion of a fortress from below.
Each action was small. Agents switch firms. Clients follow agents. Board members talk to outsiders. None of these events, taken individually, would have alarmed anyone at William Morris. But Emanuel was executing a sequence designed to compound, and compounding is invisible until it is irreversible.
The military parallel is exact. In the seventeenth century, the French military engineer Sebastien Le Prestre de Vauban perfected the art of the sap: a trench dug in a zigzag pattern toward a fortified position. The sappers worked at night, under cover, advancing a few yards each day. No single night’s progress was threatening. The garrison inside the fortress could see the trench approaching but could not stop it without exposing themselves to fire. By the time the sap reached the fortress wall, the defenders’ position was untenable. The assault, when it came, was a formality. Louis XIV won more territory with Vauban’s patient digging than with all his cavalry charges combined. The lesson is one that ambitious operators resist: the slow approach is often the most devastating, precisely because the target does not recognize the danger until the trench has already reached the wall.
Emanuel’s sap reached the William Morris wall in 2009. When the final merger vote came, the William Morris board voted to remove their own CEO and CFO before the deal even closed. A board of directors firing their own leadership in order to surrender more efficiently to a competitor: this is what seven years of silent digging looks like from the inside.
In February 2012, Mark Zuckerberg wrote an internal email that laid out Facebook’s acquisition strategy with comparable clarity. There are a finite number of social mechanics that can be invented, Zuckerberg argued. A mechanic is a specific form of interaction: sharing photos, sending messages, broadcasting status updates, creating groups. Once an outfit wins at a specific mechanic, supplanting them is nearly impossible. The winning outfit has the users, the data, the network effects, and the product iteration that comes from operating at scale.
Zuckerberg’s solution: buy the organizations that have won at specific mechanics, leave their products running, and incorporate the social dynamics they invented into Facebook’s core products. Instagram had won at photo sharing. WhatsApp had won at messaging. Each acquisition gave Facebook not just users or revenue but the crew that had figured out a social mechanic that Facebook’s own engineers had failed to develop internally. Instagram had thirteen employees when Facebook bought it for a billion dollars. The price per head was seventy-seven million dollars. No human being is worth that in salary. But the group that had iterated on the photo-sharing mechanic until they found the version that worked was worth more than any salary could capture. The knowledge was embodied in the crew. It could not be transferred through documentation or training or consulting engagements. It could only be acquired by acquiring the individuals who held it.
The Zuckerberg memo also contained an implicit admission that should unsettle every large enterprise in the world: Facebook, with its thousands of engineers and billions of dollars, could not reliably develop new social mechanics internally. The operation was too large, too focused on its existing products, too constrained by its existing user base to experiment freely. Small crews operating independently were better at inventing than large ones operating inside an established institution. Facebook was confessing that it could not build what it needed and was willing to pay a massive premium to buy it.
Two years later, the counter-story arrived. TikTok blindsided Facebook because it proved that engagement could exist on an app with no connection to your social network. The entire theory that the social graph was the central asset, that all engagement required knowing who your friends were, turned out to be wrong. An algorithm could substitute for a friend list. Zuckerberg’s acquisition thesis had a hole in it large enough for ByteDance to drive a truck through. The mechanics Facebook bought were real. The assumption they sat on top of, that social connection was the substrate of all engagement, was not. Every acquisition strategy has a hidden assumption. The assumption is invisible because the strategy’s success confirms it, right up until the moment reality stops cooperating.
In March 2019, Jensen Huang saw a different kind of acquisition opportunity. Nvidia announced it was acquiring Mellanox, an Israeli networking outfit, for seven billion dollars in cash. Intel had been considering the purchase. Nobody in the industry understood what a GPU maker wanted with a networking firm.
The answer was architectural. The unit of competition was shifting from the chip to the data center. A single GPU, no matter how powerful, was useless for training a large language model. The training required thousands of GPUs working in concert, and “working in concert” meant communicating with each other at enormous bandwidth and negligible latency. The bottleneck was not compute but communication between compute nodes. Mellanox built the InfiniBand networking technology that solved the bottleneck. Seven billion dollars purchased capability, institutional knowledge, and exclusivity in a single transaction. Intel, which might have acquired Mellanox, was left without the networking layer that would have completed its own AI offering. By the time Intel recognized the gap, it was structural.
Five months after the Mellanox acquisition, Nvidia released Megatron, an 8.3 billion parameter language model trained on 512 GPUs for nine days. The research function became a recruiting magnet: the best AI researchers wanted to work at the operation that built the best AI infrastructure, and the best AI infrastructure was built by the outfit that employed the best AI researchers. The flywheel was self-reinforcing and, for competitors, self-defeating.
Emanuel’s sap, Zuckerberg’s social mechanic acquisitions, Huang’s data center play: three different operators, three different industries, same underlying move. Identify where the capability lives. Acquire it. Deny it to competitors.
The Succession Chasm
In pre-Christian Scandinavia, the state was the king. When the king died, the networks of loyalty, gift exchange, and obligation that constituted his authority died with him. Each new king had to reinvent his kingdom from scratch: negotiate new alliances, establish new gift relationships, earn new loyalty through new victories. Christian kingdoms solved the problem by embedding succession in institutional infrastructure. The crown passed from father to son by right, not by negotiation. The bureaucracy continued to function regardless of who wore the crown. The distinction maps cleanly onto corporate life. The founder-led outfit whose authority depends on the founder’s personal relationships, personal judgment, and personal reputation is a Scandinavian kingdom. The institution whose authority depends on systems that outlast any individual leader is a Christian one.
Pharaoh Pepi II came to the throne at age six and ruled for over sixty years. During his reign, the regional governors who administered Egypt’s provinces transformed their positions from appointed offices into hereditary fiefdoms. By the time Pepi died, the governors were feudal lords who owed nothing to the new pharaoh. The central authority had been quietly hollowed out over six decades of a single ruler’s neglect. The empire fragmented within a generation of his death. Pepi holds the record for the longest reign in Egyptian history and possibly the worst return on longevity in the history of governance. Sixty years on the throne. Sixty years during which his empire was hollowed out beneath him while he presumably assumed that the stability he observed every morning from his palace window was evidence that things were fine. He is the patron saint of every CEO who looks at quarterly earnings, sees the numbers holding, and concludes that the talent exodus happening three levels below him is someone else’s problem.
When Home Depot’s founders, Bernie Marcus and Arthur Blank, stepped down in 2000, the board hired Robert Nardelli from GE to replace them. Nardelli brought GE’s discipline to Home Depot. Margins improved. The outfit passed one thousand stores. The numbers looked excellent.
But the culture collapsed. Customer service scores plummeted. Employee morale cratered. The associates who had made Home Depot famous, the ones who would drive to a customer’s house to install a chandelier after hours, were replaced by part-time workers reading scripts. Nardelli ran Home Depot the way GE ran a turbine division: through metrics, through efficiency, through cost control. He managed things. Marcus had managed humans. The difference was invisible on the balance sheet and catastrophic on the sales floor.
Nardelli resigned in 2007 with a two-hundred-ten-million-dollar severance package, paid, in effect, for proving that founder culture does not survive transplant surgery. A lesson the board could have learned for free by reading any of the fifty books on the subject. His successor, Frank Blake, did something that almost no successor in corporate history has done: he called the founders and asked them to teach him how to rebuild what they had created. The gesture was an admission of ignorance. It was also a form of acquisition: Blake was acquiring the founders’ institutional knowledge the way other CEOs acquire employees. He reinvested in training. He empowered associates. He rebuilt the culture slowly, one store at a time, one hire at a time. The recovery took years. The damage took months. The asymmetry is the lesson: culture builds slowly and decays quickly. Building a dense roster of gifted operators is the work of decades. Destroying it is the work of a single bad leader with a spreadsheet and a mandate to improve margins.[7]
When Mitt Romney left Bain Capital in 1999 to run the Salt Lake City Olympics, the firm faced the same structural choice. Most founder-oriented investment firms did not survive more than one generation. The founder’s reputation attracted capital. The founder’s judgment selected investments. The founder’s network provided deal flow. When the founder left, everything left with him.
John Connaughton and the remaining partners made a deliberate choice: flat partnership with distributed economics. No single partner would dominate. No single partner’s departure would destabilize the firm. The decision was a retention strategy wearing governance clothing. By distributing economics broadly, the firm retained partners who might otherwise have left to start their own shops. By eliminating the founder’s disproportionate share, the firm sent a message to every partner: your value here is based on what you contribute, not on when you arrived.
Bain also made a hiring decision that separated them from the rest of private equity. Seventy-five percent of Bain’s hires came from consulting firms. The rest of the industry hired almost exclusively from investment banks. The difference mattered because consultants and bankers are different animals at the molecular level. Bankers chase fees. They optimize for deal volume and transaction size. Consultants are trained to diagnose problems and design solutions. They optimize for analytical rigor and client outcomes. The operating orientation compounded over decades into a competitive advantage that transaction-oriented firms could not replicate, because the advantage was embedded in the judgment and habits of hundreds of individual partners, not in any process or system that could be copied.[5]
In 2020, Dario Amodei was VP of Research at OpenAI. He had built the research group. He had recruited many of the researchers. He had shaped the research agenda.
He left. He took several key researchers with him. They founded Anthropic.
The departure demonstrates what Thomas Schelling called a “focal point” in coordination games. When a group of researchers must decide independently whether to stay at an institution or follow a departing leader, they do not need a recruiting campaign. They do not need phone calls or offers or persuasion. They coordinate around the most salient signal. Amodei’s departure was that signal: the event around which individually uncertain researchers could coordinate a collective exit without explicit negotiation. OpenAI did not lose its researchers to a recruitment effort. It lost them to a coordination event that required no effort at all. Three years later, in November 2023, the dynamic replayed at grotesque scale: when OpenAI’s board fired Sam Altman, over seven hundred of the company’s roughly eight hundred employees signed a letter threatening to follow him to Microsoft. The Schelling point did not need Altman’s encouragement. It barely needed twenty-four hours. The concentration of capability that institutions assemble at enormous expense, over years of careful recruitment, can walk out in a week. Sometimes in a weekend.
The Roman army understood this risk two thousand years ago. Rome renewed its legions every fifteen years. Professional soldiers who served indefinitely developed loyalties to their generals rather than to the state. The general who commanded a loyal legion for twenty years had a personal army. The rotation prevented any general from building a fiefdom within the institutional army. In the late Republic, Gaius Marius reformed the army by allowing landless citizens to enlist. The reform solved the immediate problem of recruitment but created a structural one: soldiers who owned no land depended entirely on their general for their livelihood. Within a generation, Roman generals were marching their personal armies against Rome itself. The loyalty that had been the Republic’s strength became the instrument of its destruction.
Amodei later articulated his own philosophy: “It is incredibly unproductive to try and argue with someone else’s vision. Take some people you trust and go off together and make your vision happen.” The statement is both a recruiting pitch and a warning to every employer in the room. If your best researcher’s loyalty belongs to a person rather than to your institution, that person owns your roster. You just pay the salaries.
The Eye Test
Michael Ovitz built Creative Artists Agency into the most powerful agency in Hollywood by applying a method that sounds simple and is almost impossible to replicate.
“Every Sunday for fifty years, I do the same exercise,” Ovitz said. “Hell or high water, I’ve never not done it. I look at my calendar the week before. I go through every single meeting I had, every transaction, every human I met, could be social. And I decide if they go on what’s called the Sunday list.”[4]
Fifty years. Twenty-six hundred consecutive Sundays. Each producing a list of relationships to cultivate, connections to deepen, opportunities to explore. The list compounded. The relationships compounded. The information that flowed through those relationships compounded. By the time Ovitz was at the peak of his influence, he did not need to seek out information. Information sought him out, because he was the node through which the most relationships passed. The Sunday list was a compounding engine with a fifty-year time horizon, and the returns, like all compounding returns, were invisible for the first decade and overwhelming by the fourth.
Ovitz’s method for evaluating talent was borrowed from Scorsese. The eyes reveal what the resume conceals: hunger, intelligence, attention, presence. The individual who looks at you and is already thinking about what to say next is a different creature from the one who looks at you and is absorbing what you are saying. The second learns. The first performs. Organizations filled with performers look impressive in the lobby and accomplish little in the conference room. Organizations filled with learners look modest and accomplish everything.[4]
Ovitz inverted the standard hiring approach. Most firms assess candidates by examining past achievements: schools attended, positions held, revenues generated. Ovitz assessed candidates by examining their capacity for future learning. A mind that learned quickly and continuously would, given enough time, surpass the credentialed achiever with a fixed skill set. The compounder would overtake the achiever. The race was always won over decades, never at the starting line.
Andrew Carnegie practiced a different, more clinical version of the eye test. In April 1876, Carnegie wrote a letter that outlined, with what his biographer called “an astonishing, almost uncanny exhibition of clairvoyant faculty,” the personnel changes that would occur at his firm over the following six years. Every change he predicted came to pass.
The prediction was analytical, not psychic. Carnegie studied his operators the way he studied his cost sheets: methodically, continuously, with an attention to detail that bordered on surveillance. He knew who was growing and who was stagnating. He knew who would thrive under pressure and who would crack. Carnegie’s method was Bayesian in spirit. He started with an initial assessment of each individual. He updated that assessment continuously with new evidence: observed behavior under varying conditions, responses to pressure, capacity for growth. He arrived at a prediction about future behavior that was more accurate than any snapshot assessment could produce. The modern interview is a single data point. Carnegie’s method was a time series. The predictive power of a time series versus a single data point is categorically different, the way a film is categorically different from a photograph. A photograph can deceive. A long enough film cannot.
The implication is that the most valuable assessment tool is time. The organization that promotes from within, that advances operators who have been observed for years, has an informational advantage over the one that hires from outside based on a two-hour conversation and three phone calls. The insider has been tested by reality. The outsider has been tested by performance.
Ovitz grasped a deeper truth about how power flows in talent markets: control goes to whoever controls both sides. Having all the clients was insufficient. You needed the buyers too. When CAA represented both the actors and the directors and had relationships with every studio head, Ovitz became the unavoidable nexus of every major deal. The agency was a market-making business, never a service business. The roster was the inventory. The deals were the product. The relationships were the infrastructure. Anyone who has studied how market makers operate in finance, where the spread between bid and ask is the profit, recognizes the structure immediately. Ovitz built the Hollywood equivalent of the New York Stock Exchange, and he collected the spread on every transaction that passed through it.
The Accommodation Principle
In the summer of 2013, Mark Zuckerberg and Mike Schrep set out to recruit the individual who would build Facebook’s artificial intelligence research lab. They wanted Yann LeCun, a professor at NYU who had been working on neural networks for decades, long before neural networks were fashionable.
LeCun laid out his conditions. He lived in New York and would not move to Silicon Valley. He taught at NYU and would not quit teaching. He was committed to open-source research and would not work on proprietary technology. He wanted his group to do focused research, not the diffuse blue-sky work that characterized Microsoft Research.
Zuckerberg and Schrep accepted every condition without negotiation.
The acceptance was itself a recruiting strategy, and a devastatingly effective one. Most corporate recruiters approach conditions as obstacles to be negotiated away. The candidate wants remote work? Negotiate for hybrid. The candidate wants to continue teaching? Negotiate for one semester per year. The candidate insists on open source? Negotiate for delayed publication. Each round of negotiation communicates the same thing: we value our preferences more than we value you. Zuckerberg’s total acceptance reversed the message. LeCun did not have to wait for committee review, board approval, or compensation analysis. He was told yes immediately, to everything. The speed communicated that the decision had already been made and the conditions were irrelevant to it.
The LeCun recruitment produced Facebook AI Research, which became one of the most productive AI labs in the world. The open-source commitment that might have seemed like a concession turned out to be a competitive advantage: by publishing research openly, FAIR attracted researchers who wanted their work to have maximum impact, which attracted better researchers, which produced better research. The condition that looked like a constraint became a flywheel. This is the pattern that negotiation-obsessed recruiters miss. The condition that an exceptional hire insists upon is usually a tell about what makes them exceptional. Accommodating the condition is investing in the specific quality that makes the hire worth making.
The Qi Lu story makes the same point through a more radical gesture. In the early 2010s, Microsoft needed a leader who understood both search and artificial intelligence at a deep technical level. Qi Lu was at Yahoo, running search technology. Microsoft’s leadership group, including Steve Ballmer and Satya Nadella, flew to California for the meeting.
The meeting went well. Qi left the room. What happened next revealed something about Nadella that Ballmer has talked about ever since. Someone proposed a radical solution: hire Qi and have Nadella report to him.
“Within fifteen minutes,” Ballmer recalled, “we called Qi back. It’s what it told me about Satya. He’ll prioritize that. He doesn’t have an ego that gets in the way.”[9]
In most outfits, the decision about who reports to whom is a status competition. The existing executive defends his position. The incoming executive negotiates for rank. The resulting org chart reflects political power, not capability optimization. Nadella’s willingness to subordinate himself communicated something that presentation slides and strategic memos cannot: I care more about having the right group than I do about where I sit in it. The gesture was credible because it was costly. Anyone can say they prioritize excellence. Nadella demonstrated it by offering his own rank as the price of admission.
When the Microsoft board chose Nadella as CEO in 2014, the Qi Lu story was part of the reason. A man who would volunteer to report to his own hire was a man whose priority was the institution’s capability rather than his own title.
Consider how this plays out in your own hiring. How many conditions have you negotiated away from a candidate you wanted? How many of those negotiations sent the message that your policies mattered more than their contribution? The accommodation principle says: if the hire is exceptional, say yes to everything and figure out the logistics later. The logistics are solvable. The hire is not replaceable.
The Structural Signals
Reed Hastings told a story about a CEO at a startup where he worked early in his career. Hastings came into the office early one morning. The janitor had not cleaned up the previous night. His coffee mugs were sitting dirty on his desk. The CEO was already there. He saw the mugs. He picked them up and washed them himself.
The gesture lasted sixty seconds. The loyalty it created lasted a career. Hastings saw a leader who did not consider himself above mundane work, who did not wait for someone else to solve a small problem, who treated the office as his own space rather than as a stage for his authority. The cup-washing CEO communicated more about his leadership philosophy in sixty seconds than most leaders communicate in sixty presentations.
Every action a leader takes broadcasts to every individual in the organization about what kind of place this is. The leader who washes cups broadcasts humility. The leader who ignores dirty cups broadcasts hierarchy. The leader who complains about the janitor broadcasts blame-shifting. Each broadcast attracts operators who share the values it expresses and repels those who do not. Over years, the accumulated broadcasts create a culture that selects for a specific type of mind, and the specific type reinforces the broadcasts. The process is self-organizing. No policy required. No HR intervention necessary. Just the relentless compounding of small visible actions into institutional character.
When Robert Woodruff took over Coca-Cola in the 1920s, he articulated a hiring philosophy in four words: “Only give me the discontented.”
The line sounds paradoxical. Most firms seek content employees: satisfied, stable, reliable, unlikely to cause disruption. Woodruff wanted the opposite. He wanted operators who looked at the current state of things and saw inadequacy. The discontented employee is, by definition, the one who will change things. An organization staffed entirely by contented souls will look the same in ten years. One staffed with the discontented will look different, because the discontented will not tolerate stasis. And Woodruff’s principle works as a retention mechanism too: the discontented individual stays only as long as the firm provides enough scope for improvement. An outfit that tolerates mediocrity will lose its discontented, because there is no greater torment for a perfectionist than being surrounded by colleagues who think “fine” is a standard. Paradoxically, the discontented individual in an excellent firm is the happiest one in the building, because the pursuit of perfection is the only activity that the permanently dissatisfied find satisfying.
Judy Faulkner understood signals at a systemic level. She posted ten commandments in every bathroom and break room at Epic Systems, the health records enterprise she built from a seventy-thousand-dollar investment into one of the most dominant players in American healthcare technology. Number one: Do not go public. Number two: Do not acquire or be acquired. Number six: Focus on competency, do not tolerate mediocrity.[10] The individual who reads “do not tolerate mediocrity” on the bathroom wall and feels inspired is the individual Epic wants. The one who reads it and feels threatened is the one Epic does not want. The commandments function over time as a selection pressure, attracting minds that share the values and repelling those that do not.
Epic also inverted the typical corporate hierarchy of customer service. Every single customer had their own dedicated technical specialist crews for every product, plus a dedicated “BFF,” a single individual within Epic whose sole job was to make that customer successful. The investment required enormous human capital: thousands of dedicated specialists, each deeply embedded in a single customer’s operations. The investment paid off through lock-in. A hospital that had spent five years working with its dedicated Epic crew, that had configured every workflow around Epic’s software, that had trained every nurse and doctor on Epic’s interface, was not switching to a competitor because a competitor offered a better price. The switching cost was measured not in dollars but in institutional knowledge: the accumulated understanding of how this particular hospital operated, stored in the heads of the Epic specialists who worked with that hospital every day. Better hires create deeper relationships. Deeper relationships create higher switching costs. Higher switching costs create more predictable revenue. More predictable revenue funds better hiring. Each element amplifies every other element. Faulkner did not just invest in her workforce. She made her workforce the product, and then she made the product impossible to replace.[10]
When Louis Vuitton and Moet Hennessy merged to form LVMH in 1987, the two leaders of the combined entity, Henri Racamier from Louis Vuitton and Alain Chevalier from Moet Hennessy, were supposed to share power. The arrangement lasted until Racamier printed stationery.
The stationery had both men’s names. Racamier’s was on top. Chevalier responded by gathering every piece of the offending stationery and destroying it, which is the most expensive response to a typographical grievance in business history. The two men began fighting in the press. Their staffs chose sides. The organization fractured along loyalty lines. Neither man would yield on the question of whose name appeared first on a piece of paper that most customers would never see and that Bernard Arnault, watching from the outside, was already planning to replace with his own.[11]
While Racamier and Chevalier fought over letterhead, Arnault was accumulating shares. He had no stake in the dispute. He had capital, patience, and the ability to see that two leaders fighting over status would destroy enough organizational value to create an acquisition opportunity. LVMH today is the world’s largest luxury goods conglomerate. Arnault controls it. Neither Racamier nor Chevalier is involved. Two men of enormous ability, undone by a dispute over font placement, while a third man of comparable ability and superior patience collected everything they dropped.
In the early 2000s, the two most important individuals at Microsoft stopped speaking to each other for a year. “Bill and I went through a year where we didn’t speak,” Steve Ballmer recalled. “I didn’t know what it meant to be his boss, and he didn’t know what it meant to work for me.”[9] Two brilliant minds who cannot work together have zero combined brilliance, because the friction between them neutralizes both. Ego is the most dangerous variable in organizational design. The group of A-minus players who collaborate well will outperform the group of A-plus players who fight over stationery. The best recruiting strategy accounts not just for individual capability but for the chemical reaction between the new hire and the existing roster.
The Density Threshold
In 2004, the Capital Group, one of the most successful investment firms in history, had a problem that most firms would have envied: too many good portfolio managers, too much institutional knowledge, and a structure that was beginning to crack under its own scale. Their solution was surgical. They divided the investment function into units of approximately one hundred operators. Each unit operated quasi-independently, with its own portfolio managers, analysts, and support staff. The units were small enough for personal trust to function. The firm was large enough for institutional resources to flow. The design captured the benefits of both scales.
The logic behind the partition traces to a neurological constraint. British anthropologist Robin Dunbar observed that the size of the human neocortex limits the number of stable social relationships a mind can maintain to approximately one hundred fifty. This is biology, not management theory. A brain can track the relationships, personalities, histories, and social dynamics of about one hundred fifty individuals. Beyond that number, relationships become impersonal, communication becomes formal, and trust becomes institutional rather than personal.
The implications for roster quality are direct. At thirty, every hire is visible. A bad hire is noticed immediately by everyone and corrected quickly. At three hundred, a bad hire is noticed only by the immediate crew. The correction takes longer because the feedback travels through managers and HR departments and review cycles. At three thousand, a bad hire can persist for years, protected by the anonymity of scale, degrading the performance of the group without anyone outside the group being aware. Scale is not the friend of quality. Scale is the enemy that quality must be deliberately defended against, every day, in every hiring decision, with no exemptions and no holidays.
David Heinemeier Hansson, the creator of Ruby on Rails and co-founder of Basecamp, ran an experiment that tested the density threshold directly. He introduced engineering managers into his outfit, the way every growing technology firm does. He gave them two years to prove their value.
“Engineering management is a necessary evil when scale breaks down,” Hansson concluded. “We tried it for a couple of years. After that I thought, no, I was right. Not every programmer needs a therapy session with an engineering manager every week.”
Hansson eliminated the engineering manager role. Programmers reported directly to the founders. Decisions were made by the individuals who wrote the code, not by those who managed them. The model works when the caliber of the roster is high enough that management is redundant. It fails when the caliber is low enough that management is necessary. The threshold between the two states is the density threshold: the point at which the average mind in the room is capable of managing itself. Below the threshold, managers are essential. Above it, managers are overhead. Most firms operate below the threshold and therefore need managers. A few, the ones that have invested decades in who they let through the door, operate above it.
Here is the thesis collision this volume owes you. Everything you have read argues that concentrating exceptional talent produces extraordinary outcomes. The IDA mathematicians. Netflix’s Keeper Test alumni. Lemann’s PSDs. The evidence is consistent and the mechanism is clear.
But in metallurgy, the purest metals are often the most brittle. Pure iron shatters on impact. The impurities in steel, the carbon, the chromium, the manganese, are what give it tensile strength. The alloying elements absorb stress that the pure metal cannot. The organizational parallel is uncomfortable: the “impurities” that density-obsessed outfits remove, the merely competent, the steady-but-unspectacular, the operators who scored B-plus on the Keeper Test, may be the organizational equivalent of carbon in steel. Their presence absorbs shocks that the A-plus players cannot absorb alone, because the A-plus players are optimized for performance, not resilience. The most decorated special forces units in military history still carry medics, logistics specialists, and communications officers who would fail the combat fitness standards applied to the operators. The support structure is not the mission. But without it, the mission fails.
The more you concentrate capability around a small number of exceptional individuals, the more catastrophic the loss of any one of them becomes. An operation of a thousand adequate employees can lose fifty and barely notice. An operation of fifty exceptional ones can lose five and face an existential crisis. The talent raid thesis says density is everything. The talent raid evidence says density is also the single greatest source of organizational risk. Both are true. The author does not have a tidy resolution. The Romans tried rotation. Bain tried distributed economics. Netflix tried institutional loyalty to a standard. None of these solutions fully resolves the contradiction. They just manage it, imperfectly, at a cost. The question every density-obsessed leader must answer: where does your organization’s carbon come from, and have you already removed it?
The Operator’s Toolkit
The conventional wisdom about building exceptional groups fits on a motivational poster. Hire A-players. Create a great culture. Pay competitively. Empower your workforce. You have heard this advice. You have probably given it. The reason it fails is that it operates at the level of aspiration rather than mechanism. It tells you what to want without telling you what to build. The operators in this volume did not succeed because they wanted great talent more than their competitors did. They succeeded because they built structural systems that produced great rosters whether or not any individual manager was paying attention on any given Tuesday.
The following five practices are derived from operators who actually built and maintained organizations of extraordinary quality, and each one addresses a specific failure mode that no amount of inspirational leadership could have prevented.
The Hastings Inversion. When Reed Hastings analyzed Pure Software’s decline, he did not ask “how do we hire better?” He asked a different question: what systemic conditions caused our best to leave? The answer was that rules created to compensate for mediocre hires drove out the excellent ones. The process manual that protected the outfit from the bottom twenty percent was experienced as an insult by the top twenty percent, and the top twenty percent were the ones with options.
The inversion operates at the level of organizational infrastructure, not individual judgment. Every quarter, take your roster’s complete process inventory: every standing meeting, every approval chain, every review cycle, every required report. For each one, ask two questions. First: if everyone on this roster were excellent, would we still need this? Second: who specifically is this designed to catch? If the answer to the first is no and the answer to the second is “the one or two operators who make mistakes,” you have identified a process that exists to manage mediocrity. The cost of that process is not the time it consumes. The cost is the signal it sends to the fourteen who never needed it: we do not trust you enough to let you work.
The implementation step most leaders skip: after identifying the process, do not announce its removal. Remove it silently for one quarter and measure what happens. If nothing breaks, the process was insurance against a risk that no longer exists. If something breaks, you have identified an individual, not a process problem, and you now have evidence for a conversation you should have had months ago.
The failure signature is specific: you conduct the audit and conclude that every process is necessary. If that is your conclusion, you have not conducted the audit. You have defended the status quo. An honest audit always identifies at least one process that exists because firing someone was harder than adding a review step. The review step is still there. The individual may have left years ago. The scar tissue remains.
The Lemann Filter. What does it cost your organization when a bad hire survives six months? Calculate the number. Salary, management overhead, opportunity cost of the seat, damage to the group’s morale. Now multiply that by the number of bad hires who survived six months in the last two years. That number is the price of your current filter.
Lemann built a compensation architecture that made traditional evaluation unnecessary. Below-market base salary. Extreme performance bonuses. Mandatory equity purchase with personal funds. The system does the sorting: operators who believe in their own ability opt in. Those who want safety opt out. Before you look at a single resume, design your compensation, your work environment, and your cultural signals so that the kind of hire you want is attracted and the kind you do not want is repelled.
Write down the profile of the individual who would hate working at your outfit. Be specific. Not “lazy.” Lemann could describe his anti-candidate in four words: comfortable, credentialed, risk-averse, entitled. What are your four words? Now examine every touchpoint a candidate encounters: your job posting, your office, your interview process, your offer letter, your first-week onboarding. At each touchpoint, ask: would my anti-candidate find this appealing or repellent? If appealing, you have a leak in your filter. Your job posting that emphasizes “competitive salary, comprehensive benefits, and work-life balance” is a beacon for the exact applicants Lemann was trying to exclude.
The filter must also operate on your existing roster. Every time you add a perk, raise base salaries faster than performance bonuses, or make equity grants free instead of purchased, you are diluting the filter. The roster that joined under the original terms was self-selected for hunger. The roster that stays under softer terms is self-selected for comfort. The transition happens gradually enough that no single decision triggers alarm. By the time the founder notices the change in the room’s energy, the operators who would have thrived under the original terms have already left for somewhere that still demands what they are willing to give. One exception to the filter is an anomaly. Two exceptions are a pattern. Three exceptions and you no longer have a filter. You have a suggestion.
The Ovitz Calendar. Consider two candidates for a senior role. One you met last week. She interviewed well, her references checked out, and she presents as brilliant. The other you have watched for four years, through two market downturns, a product failure, and a leadership transition. She was not always brilliant. But she grew faster than anyone you have observed, and the growing never stopped. Which hire carries more risk?
Ovitz’s Sunday list practice, maintained without exception for fifty years, demonstrates that relationship capital builds the same way financial capital does: through small, consistent deposits that compound over decades. Block ninety minutes every Sunday. Open your calendar from the past week. For every meeting, every call, every casual interaction, ask three questions. First: did I learn something from this individual that I could not have learned elsewhere? If yes, they go on the deepen list. Second: did this individual demonstrate a capability that surprised me? If yes, they go on the watch list, a roster of minds whose trajectory you want to track over years, not months. Third: is there someone I met through this individual who I should know independently? If yes, they go on the connect list.
The deposits are small. A follow-up email with an article relevant to something they mentioned. An introduction to someone in your network who could help them. A note six months later referencing the specific thing they said that stuck with you. None of these takes more than five minutes. All of them compound. The operator who receives a thoughtful follow-up six months after a casual conversation remembers you differently from the one who receives a LinkedIn connection request thirty seconds after a handshake.
Carnegie predicted his firm’s personnel changes six years in advance because he had observed his operators with the patience of an anthropologist. You cannot replicate his predictive accuracy from a two-hour audition. You can replicate it with five years of Sunday lists. The best hire you will ever make is a mind you have watched for years. The worst hire you will ever make is the impressive stranger. If your Sunday list becomes a chore you complete mechanically, adding names without deepening relationships, you are maintaining a database, not building capital.
The Ford Wage Signal. In 2024, a mid-stage startup founder told me she had lost her best engineer to a competitor offering fifteen percent more base salary. The engineer had not asked for a raise. He had not signaled unhappiness. He simply got a better offer and left. The founder’s question was: how do I prevent this? The better question was: what had she broadcast over the previous eighteen months that told the engineer he should be listening to outside offers in the first place?
Ford’s five-dollar day was a broadcast, and the broadcast was more valuable than the wage. Identify the single most visible signal your outfit sends about how it values its talent, and make that signal dramatically louder than your competitors’. Start with a diagnostic. List the last five things your organization did that a current or prospective hire would have noticed. Not the things you announced in an all-hands meeting. The things that traveled through the informal network: the story someone told over lunch, the decision that made operators update their assumptions about what kind of place this is. Now categorize each one. Did it signal investment or extraction? Trust or control?
For Ford, the costly signal was wages. For Zuckerberg recruiting LeCun, it was total accommodation. For Nadella recruiting Qi Lu, it was the willingness to restructure the org chart around a single hire. The specific signal matters less than two properties it must have. First, it must be visible: a decision that travels through the informal network within twenty-four hours. Second, it must be costly. A signal that costs you nothing communicates nothing. A signal that costs you something substantial, money, status, organizational convenience, the founder’s ego, communicates everything.
And the signal must be repeated. A single costly gesture creates a story. Repeated costly gestures create a reputation. Ford did not raise wages once. He maintained the premium for decades. When your outfit makes one dramatic gesture and then reverts to normal, the broadcast is: we did that once for PR. When it makes the gesture repeatedly, absorbing the cost each time, the broadcast is: this is who we are. What has your outfit sacrificed, visibly and recently, to attract or retain a specific hire? If you cannot name an example from the last six months, your signals are cheap. Cheap signals attract no one worth having.
The Bain Distribution. Every founder-led organization carries a ticking structural bomb that the founder cannot hear because the ticking sounds like loyalty. The bomb is this: the founder’s disproportionate share of economics creates resentment that eventually drives out the operators doing the work. The resentment does not arrive on day one. It arrives on the day a partner calculates that their contribution exceeds their compensation, and that the gap flows to someone whose contribution is now smaller than theirs. That day always comes. The only question is whether the economic structure acknowledges it before the departures begin.
When Bain Capital chose flat partnership economics over founder monetization, they were defusing the bomb while it was still quiet. Map the economic structure of your group, division, or organization. For every operator in a senior or critical role, answer three questions. First: what percentage of the value they create do they capture? Second: what percentage does someone above them capture from their work? Third: does the individual in the senior role know the answer to the first two questions? They almost certainly do. Operators are better at compensation math than leaders want to believe.
Now run the Bain test: if your three best operators left tomorrow and started a competing firm, would your economic structure be one of the reasons they cite? If the answer is yes, you have a structural vulnerability that no amount of culture, mission, or camaraderie will permanently offset.
Redistribution must happen before the resentment surfaces. Once resentment is visible, redistribution looks like a concession, and concessions breed contempt. Bain distributed economics while Romney was still present and still dominant. The distribution felt like generosity, not capitulation. By the time Romney left, the structure was already built and the partners were already invested in the collective model. Had Bain waited until Romney’s departure to redistribute, the gesture would have read as desperation. The timing of the redistribution matters as much as the redistribution itself. The failure mode: you redistribute economics but retain disproportionate control. This is worse than doing nothing, because it sends a contradictory message. We trust you enough to share the money but not enough to share the decisions. The Bain model distributed both. Distribute one without the other and you have created a new resentment that is harder to name and harder to fix.
The Limit of the Thesis
In February 2024, Kraft Heinz wrote down $1.2 billion in goodwill, the latest in a cascade of impairments that had destroyed over twenty billion dollars in shareholder value since the 2015 merger. The playbook that built the world’s largest beer empire had been applied, with identical confidence, to a portfolio of food brands. The result was gutted R&D, declining brand equity, and a stock price that had lost half its value. Lemann’s PSD system, the most effective talent filter in the history of consumer goods, had produced a workforce that was superb at cutting costs and incapable of imagining a new product. The system that filtered for hungry, risk-tolerant operators also filtered out the minds who might have said: this time is different.[6]
It would be satisfying to end with five clean practices and the implication that applying them will build an unstoppable roster. But the evidence in this volume has one more property that an honest essay cannot ignore: the operators who executed these practices most successfully also produced the most spectacular failures.
Hastings built Netflix’s culture of radical candor and lost twenty percent of his workforce in year one. Some of those departures were the system working as designed. Some were gifted individuals who left because the environment felt brutal, not rigorous, and the line between those two things is thinner than any Keeper Test can measure. Facebook’s acquisition strategy captured Instagram and WhatsApp and missed TikTok. Carnegie predicted his firm’s personnel changes six years in advance and also presided over the Homestead Strike, where his cost-cutting approach to labor produced a battle that left ten dead. The eye for evaluating subordinates and the blindness toward the workers those subordinates managed coexisted in the same man, because the same trait, relentless optimization of the roster at the top, can produce both extraordinary capability and extraordinary cruelty depending on where you draw the boundary of “the roster.”
Every practice in the toolkit has a failure mode that the practice itself cannot detect. The Hastings Inversion can remove process that was actually load-bearing. The Lemann Filter can select so aggressively for one type of mind that the firm becomes a monoculture, brittle in exactly the ways that monocultures always are. The Ovitz Calendar can become a tool for accumulating influence rather than insight, and the line between the two is not always visible from the inside. The Ford Wage Signal can attract mercenaries who love the broadcast more than the work. The Bain Distribution can eliminate the founder’s ability to make unpopular long-term decisions, producing an outfit that optimizes for partner consensus rather than for excellence.
Mary Kay Ash discovered this when her override commission system, the multiplicative income structure that turned a tiny percentage into serious money through network effects, was destroyed in a single administrative decision at her previous employer. She had built the network. She had trained the recruits. She had created the compounding income stream. Then the employer changed her territory and reassigned everything she had built. When she founded her own organization, she designed permanent territories and permanent override relationships specifically to prevent the same vulnerability. She solved the problem. But she solved it only because she had first been destroyed by it. Every structural innovation in this volume was forged in the same fire: an operator encountered a failure so specific and so personal that they designed a system to prevent it from ever happening again. The system worked. Until it didn’t.
The honest conclusion is that these practices will make you better than the alternative, which is no practices at all, and that the gap between “better than the alternative” and “saved” is where the real work of leadership lives. The roster you build will be the best thing about your organization and the most fragile thing about it. The two properties are inseparable.