The Best and the Brightest, and Still Wrong

Why intelligence, expertise and an impressive board are not enough to lead a company.

The disturbing paradox of intelligent failure

When John F. Kennedy entered the White House in 1961, he surrounded himself with what seemed to be an unbeatable concentration of intellectual talent. His advisers came from Harvard, Wall Street, the military and the commanding heights of American industry; they were articulate, analytical and supremely self-confident, and they shared their president’s conviction that intelligence, rationality and modern methods of management could solve problems that had defeated less sophisticated administrations.

Lyndon Johnson, Kennedy’s vice-president, was reportedly dazzled by them. Sam Rayburn, the veteran Speaker of the House and one of the most experienced political operators in Washington, was less reassured. After listening to Johnson describe the brilliance of the new administration, he observed that he would feel rather better if just one of these men had once run for county sheriff.

Rayburn was not defending ignorance against intelligence. He was pointing towards a distinction that is as important in business as it is in politics: the difference between intelligence as an individual faculty and judgement as a quality developed through contact with reality. The first can be demonstrated in an examination room, a laboratory or an investment committee; the second is acquired through exposure to consequences, conflicting interests, incomplete information and human beings who stubbornly refuse to behave as models predict.

In The Best and the Brightest, David Halberstam examined how Kennedy and Johnson’s exceptional advisers progressively involved the United States in Vietnam. The enduring interest of the book lies in the fact that its protagonists were neither fools nor obvious ideologues. They were serious, accomplished and, within their respective domains, highly competent men. Their failure was therefore considerably more unsettling than simple incompetence: they proved capable of deploying immense intelligence in defence of a mistaken representation of the world.

They interpreted Vietnam largely through the framework of the Cold War and the domino theory, reducing a conflict shaped by nationalism, colonial history, local legitimacy and Vietnamese political culture to another move in a global contest with communism. Once this model had been adopted, information was sorted according to whether it confirmed or disturbed it. Diplomats and observers with direct knowledge of the country were marginalised; quantitative indicators and military reports acquired an authority that their underlying assumptions did not merit; setbacks were interpreted not as evidence that the strategy might be mistaken, but as proof that it had not yet been applied with sufficient determination.

The result was a familiar escalation mechanism. Each additional commitment made the previous one appear more important, while the political cost of admitting error rose with the resources already invested. A policy that had never been chosen in a single, fully considered decision gradually became almost impossible to reverse. The administration repeatedly sought to “keep its options open”, yet every attempt to postpone a fundamental choice reduced the range of good options actually available.

Frontier tech companies should recognise themselves in this story, because they, too, deliberately assemble unusual concentrations of intelligence around difficult and uncertain problems. Their founders may be eminent professors, successful entrepreneurs or former executives from major industrial groups. Their boards include accomplished investors, scientists, public officials and corporate leaders. Their presentations contain patents, simulations, technical milestones, market forecasts and elaborate risk matrices, all of which convey an impression of disciplined mastery.

Nevertheless, many of these companies make decisions that appear, in retrospect, astonishingly detached from the evidence available inside their own organisations. They continue financing programmes after the scientific premise has weakened; they announce industrial timetables that their manufacturing teams regard as implausible; they treat regulatory objections as communication problems, customer resistance as a failure of education, and recurring technical anomalies as isolated incidents. Because the people involved are highly intelligent, they are rarely short of convincing explanations for doing so.

This is the paradox at the centre of deep-tech leadership: intelligence is indispensable, but it does not necessarily protect an organisation from error. Under certain conditions, it makes error more durable, because intelligent people are exceptionally capable of rationalising commitments, defending elegant models and producing sophisticated accounts of why contradictory evidence should not yet be believed.

Why frontier tech magnifies the problem

The expression frontier tech or “deep tech” covers very different sectors—biotechnology, semiconductors, robotics, new materials, quantum computing, aerospace, energy and advanced industrial systems—but these companies share a distinctive strategic structure. Their success depends not on one uncertainty but on a chain of interdependent uncertainties, each governed by a different body of knowledge and each capable of invalidating the economic proposition.

A scientific effect must first exist and be reproducible. It must then be transformed into an engineered system, manufactured at an acceptable yield, certified where necessary, incorporated into a customer’s operations, maintained in the field and sold at a price that supports the capital structure of the business. A company may be right about the science and wrong about industrialisation; right about the product and wrong about adoption; right about customer interest and wrong about the regulatory pathway; or right about all of these and still discover that the financing required to cross the gap between prototype and scale exceeds what its investors can provide.

No individual, regardless of ability, can master this entire chain. A distinguished physicist may understand a phenomenon better than anyone else in the world while possessing limited knowledge of quality systems, production economics or organisational design. A former executive from a global corporation may be highly effective at optimising mature operations but poorly adapted to a company in which even the fundamental technical architecture remains uncertain. Expertise, in other words, is local, whereas the CEO’s decisions are systemic. This creates an unavoidable information asymmetry: the person with the greatest authority rarely possesses the most relevant knowledge about every decision over which that authority is exercised.

In a healthy organisation, these partial views are combined into a progressively more accurate representation of reality. In an unhealthy one, they are ranked according to status: the abstraction held at the top defeats the observation made at the edge, even when the latter contains the information on which the company’s survival depends.

This is why the familiar distinction between vision and execution is particularly dangerous in deep tech. According to this managerial mythology, the leader defines the destination while the organisation handles implementation, which is often treated as a form of corporate plumbing: necessary, occasionally difficult, but fundamentally subordinate work that can be delegated once the architecture has been decided.

Yet in deep tech, the “plumbing” is precisely where the strategy encounters physics, regulation, economics and human behaviour. Whether a material can be produced consistently outside the laboratory, whether a sensor remains reliable under vibration and temperature variation, whether a biological process behaves similarly at ten thousand litres as it did at ten, whether a supplier can meet the required tolerances, or whether a customer can integrate the product without redesigning its operations are not details of execution. They determine whether the company possesses a business at all.

When technical accidents become organisational biographies

Catastrophic technological failures are usually described initially through their physical causes: a damaged heat shield, an erroneous software deployment. These causes matter, but investigations repeatedly reveal that the physical defect was only the final link in a longer organisational chain. The accident becomes, in effect, an involuntary biography of the institution that produced it.

Before the Challenger disaster in January 1986, engineers at Morton Thiokol expressed serious concern about the performance of the shuttle’s solid-rocket-booster seals in unusually cold weather. During the discussion preceding launch, Thiokol management reversed its initial recommendation and supported proceeding, contrary to the position of its engineers. The Rogers Commission subsequently identified failures of communication, a conflict between engineering evidence and managerial judgement, and a NASA structure that allowed safety issues to bypass senior shuttle managers. Challenger disintegrated 73 seconds after liftoff.

The case is often simplified into a morality tale in which engineers knew the truth and managers ignored them. The reality is more instructive. The available data were incomplete, the relationship between temperature and O-ring performance was contested, and a decision had to be made under operational pressure. This ambiguity did not reduce the need for caution; it made the quality of the decision-making process decisive. The failure lay partly in the way uncertainty was framed: instead of requiring evidence that launch was safe under unprecedented conditions, the discussion placed pressure on engineers to establish conclusively that it was unsafe.

Seventeen years later, the Columbia Accident Investigation Board concluded that the management practices overseeing the shuttle programme were as much a cause of the loss of Columbia as the foam strike that damaged its wing. Engineers had attempted to obtain better imagery of the orbiter in flight, but their concern did not produce an adequate response from the decision-making hierarchy. Schedule pressure, normalised anomalies and a fragmented safety structure again influenced the interpretation of ambiguous evidence. The board recommended an independent technical authority responsible for safety requirements and waivers, explicitly separated from responsibility for cost and schedule.

The repetition is more disturbing than either event considered separately. NASA was not an organisation that had never thought about safety; after Challenger, it had studied its failures extensively, revised procedures and pledged to learn. Columbia demonstrated that institutional learning is not permanent. Lessons decay as personnel change, commercial or political pressure returns, and practices introduced in response to a crisis become rituals whose original purpose is gradually forgotten.

An anomaly that does not produce a catastrophe can even weaken vigilance. Each successful mission in which foam was shed, or each launch in which an O-ring suffered damage without causing loss of vehicle, made the deviation appear more acceptable. The absence of disaster was misread as evidence of safety, although it might equally have been evidence of good fortune. This process, often described as the normalisation of deviance, is particularly dangerous in deep tech because systems can operate successfully for long periods while approaching a boundary that the organisation does not fully understand.

Theranos represents a more extreme and ethically different case, because its leadership crossed the boundary from excessive optimism into deception. Nevertheless, it demonstrates how prestige, secrecy and narrative control can isolate a technical company from corrective evidence. According to the SEC, Theranos raised more than $700 million while making false or exaggerated claims about its technology, business and financial performance. The company represented that its proprietary analyser could perform an extensive range of tests from very small blood samples; the SEC complaint stated that the device was used for only 12 of the tests offered to patients, with the majority performed on modified commercial equipment.

Theranos should not be used to imply that every technically overambitious founder is fraudulent. The more useful question is why an organisation operating in a field as demanding as medical diagnostics was governed in a way that allowed charisma, confidentiality and executive authority to become substitutes for independent technical verification. A prestigious board cannot compensate for insufficient domain knowledge, and loyalty to a mission cannot justify preventing qualified people from examining whether the mission’s central claims are true.

More recently, an expert panel examining Boeing’s safety culture reviewed thousands of pages of documents and interviewed more than 250 employees, managers, executives, supplier personnel and regulators. Among its observations was a particularly revealing criticism of Boeing’s “Seek, Speak & Listen” framework: the panel found considerable emphasis on speaking, but much less on seeking and listening. It also reported that the documentary and interview evidence did not provide objective support for a commitment to safety as foundational as the company’s descriptions suggested.

The distinction deserves attention. Many organisations now encourage employees to speak up, but the existence of a reporting channel says little about whether inconvenient information will influence a decision. Voice without attention is theatre; attention without authority is frustration; and authority without protection leaves the person who raises the concern dependent on the goodwill of precisely those whose plan is being challenged.

The real architecture of dissent

The usual response to these failures is to demand a better culture. Leaders announce that their doors are open, introduce whistleblowing systems, add “challenge” to the company’s values and remind employees that safety or integrity comes first. Such measures may be useful, but they often misunderstand the economics of speaking up.

For an employee, dissent is rarely costless. The benefit, i.e., preventing a future problem, is shared by the organisation and may never be visible, particularly if the warning succeeds. The potential cost (being labelled negative, obstructive, disloyal or insufficiently entrepreneurial) is immediate and personal. When the employee challenging the plan is junior, and the executive defending it controls compensation, promotion or continued employment, silence may be a perfectly rational response.

Leaders consequently learn much less from the absence of dissent than they imagine. A quiet meeting may indicate agreement, but it may equally indicate resignation, fear, fatigue or the belief that the decision has already been made. The more powerful and intellectually formidable the CEO, the more misleading apparent consensus can become, because disagreement carries both hierarchical and cognitive risk.

A functional architecture of dissent must therefore alter both process and power. For company-critical technical decisions, the person responsible for delivering the milestone should not be the sole judge of whether the evidence supporting it is adequate. Safety, technical integrity and regulatory compliance require authorities that possess sufficient independence to delay or stop a programme without being punished for the resulting effect on schedule.

The order of discussion also matters. If the CEO, scientific founder or most prestigious professor announces a view at the beginning of a meeting, everyone who follows must decide not only what they think, but whether contradicting that person is worth the social and professional cost. Asking the most junior or operationally exposed participants to speak first is not a matter of etiquette; it is a way of preventing hierarchy from contaminating the evidence before it has been collected.

Equally important is the treatment of weak signals. Employees should not be required to diagnose a complete failure mechanism before reporting an anomaly. In complex systems, the individual who notices that a measurement, sound, vibration, customer behaviour or software response is unusual may not be able to explain its significance. If the organisation demands a fully developed case before paying attention, it will systematically suppress the earliest and cheapest warnings.

The CEO should therefore be concerned with what might be called bad-news latency: the time separating the first observation of a serious problem from its arrival, in recognisable form, at the level where resources and priorities can be changed. In many failed organisations, the crucial information was not absent. It was delayed, diluted, reclassified or trapped within a part of the hierarchy that lacked the authority to act.

A doctrine of grounded leadership

The practical conclusion is not that deep tech CEOs should abandon vision, interfere in every experiment or replace expertise with instinct. Nor should organisations allow every objection to paralyse decisions, because uncertainty can never be eliminated and difficult technologies require persistence long after a conventional company would have withdrawn.

The distinction that matters is not between boldness and caution, but between conviction and dogmatism. Conviction holds that a problem is important enough to justify sustained effort and adaptation; dogmatism holds that the present explanation, architecture or timetable must be correct. Conviction permits learning because it attaches identity to the mission. Dogmatism prevents it because it attaches identity to the current plan.

The CEO must separate technical truth from organisational power. When a decision involves safety, regulatory compliance or a risk capable of threatening the company, an independent technical authority should be able to require more evidence or stop the process. That authority cannot remain meaningfully independent if career advancement depends entirely on the executive accountable for cost and schedule. NASA’s post-Columbia recommendation that technical authority be institutionally separated from programme delivery applies just as readily to a fusion company, a medical-device start-up or an autonomous-systems developer.

Capital allocation should be governed by the same discipline. When a programme misses successive milestones, management commonly responds by adding resources and urgency, partly because withdrawal would require admitting that earlier projections were wrong. Before increasing commitment, the board should require a precise account of what has been learned since the previous decision, which initial assumption has been invalidated, and why additional capital will change the underlying technical or commercial mechanism rather than merely finance another iteration of the same attempt.

This is particularly important because deep-tech progress is rarely linear. A missed milestone does not necessarily mean that a programme should be abandoned, just as a successful test does not necessarily mean that the uncertainty has been resolved. What matters is whether the company is producing information that changes the probability of success. A programme that fails intelligently may be more valuable than one that repeatedly achieves carefully selected milestones without addressing its principal risk.

Boards, for their part, must insist on proximity to evidence rather than reliance on narrative. They should periodically hear from technical, manufacturing, regulatory and commercial leaders without requiring every message to be translated by the CEO. They should examine distributions, reproducibility, yield, failure modes and confidence intervals rather than only averages and headline milestones. They should ask what evidence would disprove the investment thesis and whether the organisation is actively trying to obtain it.

The lesson of Halberstam’s “best and brightest” is therefore not that experts should be distrusted, that operational experience is always superior to analysis, or that complicated decisions should be surrendered to intuition. Deep tech companies could not exist without exceptional expertise, formal models and leaders capable of sustaining belief in achievements that initially appear improbable.

The lesson is that intelligence does not contain its own corrective mechanism. Brilliant people can construct more sophisticated reasons for ignoring contradictory evidence; homogeneous groups of experts can reinforce one another’s assumptions; detailed models can turn uncertainty into an illusion of precision; and an inspiring vision can become a means of avoiding an inconvenient present.

The decisive quality of deep-tech leadership is consequently not omniscience, but epistemic discipline: the ability to distinguish what the company knows from what it hopes, to recognise where one form of expertise ends and another begins, to remain close enough to operations that weak signals are still visible, and to revise a cherished belief without experiencing that revision as a personal defeat.

The company does not require a leader who is always right. It requires a leader who has made it institutionally difficult for everyone to remain wrong.