Part 2 of The Future of Expertise in the Age of AI. Read Part 1: When Knowledge Stops Being a Moat.
Imagine asking a surgeon how long it would take to teach you surgery. “Ten minutes,” the surgeon replies. You are surprised. Why, then, does becoming a surgeon take ten years? The answer comes back: “That is the time required to tell you what to do when something goes wrong.”
The story captures a distinction that will become increasingly important in an AI-dominated economy. Knowing the steps of a procedure is not the same as possessing the judgment to handle complications. A task may be easy to explain and difficult to master because reality does not always follow the instructions.
AI is changing the economics of this distinction. It can compress hours of intellectual work into minutes, produce sophisticated outputs with minimal human effort, and give individuals access to capabilities that once required entire teams. But as execution becomes abundant, a new question emerges: who can recognise when the system is going wrong, understand why it is wrong, and know what to do next?
The answer may help explain the emergence of a new competitive advantage: the Pilot-Mode Specialisation Moat.
1. When Execution Becomes Abundant
For much of economic history, specialised knowledge was valuable partly because it enabled people to perform tasks that others could not. A programmer could build software, a designer could create a visual identity, and a financial analyst could interpret complex data. Expertise was closely connected to the ability to execute the work.
AI is weakening that connection. A person with access to advanced tools can now generate code, analyse documents, create illustrations, and explore technical problems without possessing the same practical skills that would previously have been required. The quality of the results varies, but the cost of attempting sophisticated work has fallen.
When many people can produce similar outputs using similar tools, the ability to perform a routine task becomes a less reliable source of differentiation. The advantage shifts toward knowing which task should be performed, whether the result is correct, and how to improve it.
This does not mean that expertise has become irrelevant. It means that its economic value may migrate. The ability to produce an answer is becoming less distinctive when machines can produce thousands of answers. The ability to identify the right answer, recognise a dangerous mistake, and determine what should happen next may become more important.
The distinction is between having access to intelligence and knowing how to use it reliably.
2. The Ten-Minute Surgeon and the Ten-Year Expert
The surgeon’s answer is deliberately provocative. If a procedure can be explained in ten minutes, what accounts for the remaining years of training? The difference lies in the distance between knowing what should happen and knowing what to do when it does not.
A surgeon must recognise complications, interpret incomplete information, respond to unexpected changes, and make decisions under pressure. The procedure is only one part of the expertise. The ability to handle deviations is what makes that expertise reliable in the real world.
AI introduces a similar distinction into intellectual work. It may generate a software solution quickly, but what happens when the solution works in a demonstration and fails in production? It may analyse a financial situation, but what happens when the underlying assumptions change? It may produce a convincing scientific explanation, but what happens when the evidence contradicts it?
In each case, the problem is not simply producing an output. It is recognising when that output should not be trusted and determining the appropriate response. AI can help with these tasks, sometimes exceptionally well, but its reliability must be evaluated in the situations where errors matter.
The surgeon analogy does not imply that AI cannot acquire diagnostic or contingency capabilities. It establishes a more fundamental point: the ability to execute a procedure and the ability to manage its failure are different capabilities, and they must be evaluated separately.
3. The Pilot Is Not the Aircraft
Consider an aircraft operating with advanced automation. The pilot does not manually perform every operation. Automated systems manage much of the routine work, while the pilot monitors the aircraft, manages the flight, and responds when circumstances require intervention.
A person might learn to operate an autopilot relatively quickly. Becoming a professional pilot, however, requires much more: understanding the aircraft, interpreting its behaviour, anticipating risks, and responding to abnormal situations. Automation changes what the pilot does, but it does not automatically eliminate the need for the pilot’s wider competence.
The analogy becomes powerful when applied to AI. Imagine an expert using an advanced AI system to complete a project that previously required a team of ten people. AI conducts research, produces drafts, generates designs, performs calculations, and checks intermediate results. The expert establishes the objective, evaluates the important decisions, and intervenes when the work deviates from the intended direction.
The expert no longer needs to perform every operation personally. Instead, the expert directs a system whose execution capacity may be far greater than their individual capacity.
This is the transition from executor to navigator. The human’s value increasingly comes from directing the work rather than personally performing every step.
4. The Checkpoint Principle: Where Human Expertise Matters
The pilot analogy leads to a more precise question: if AI can automate an entire crew, where exactly does the human’s contribution remain valuable?
One answer is at critical checkpoints.
A complex workflow contains many routine operations, but not every operation carries equal importance. Some decisions are reversible and inexpensive. Others determine the direction of the entire project, expose an organisation to serious risk, or create consequences that cannot easily be undone.
An AI-assisted software project, for example, might involve hundreds of routine coding operations. Yet a handful of architectural decisions, security checks, and deployment approvals may determine whether the finished system is reliable. In medical work, automated analysis may assist clinicians, but consequential treatment decisions require appropriate clinical oversight. In research, AI may generate many hypotheses, while experimental design and independent verification determine which claims deserve acceptance.
The human does not necessarily need to supervise every action. The challenge is to identify the points where an incorrect action could invalidate the larger process.
This produces three responsibilities: deciding which checkpoints matter, determining what evidence must be examined at each checkpoint, and recognising when the system should continue, change course, or stop.
The economic significance is easy to underestimate. A decision that takes five minutes may influence the outcome of hundreds of hours of automated work. The time spent on the decision is small; the value at stake may be enormous.
This is why the quantity of human work performed may become a poor measure of its economic contribution. The more important measure may be the consequences of the decisions that human judgment controls.
A checkpoint is useful only if it can detect the relevant problem. Some errors are subtle, and an inexperienced supervisor may approve a plausible but incorrect result. Effective oversight requires genuine domain understanding, suitable verification methods, and safeguards proportionate to the consequences of failure.
5. The Pilot-Mode Specialisation Moat
Imagine two professionals with access to exactly the same advanced AI tools.
The first knows how to operate the system. They can write prompts, request revisions, and generate sophisticated outputs. The second has the same tools but also possesses deep domain knowledge. They understand the constraints of the problem, recognise when an answer is plausible but wrong, and know which assumptions must be tested before the result can be trusted.
Both have access to AI. Yet their ability to convert its output into reliable results may differ considerably.
The second professional possesses a potential specialisation moat. The moat does not arise merely from knowing more facts or using more elaborate prompts. It arises from the ability to direct intelligence toward a worthwhile objective and evaluate the resulting work.
This distinction matters because access to AI tools is likely to become increasingly widespread. If everyone can generate code, produce designs, or analyse documents, tool access alone cannot sustain a durable competitive advantage. What matters is whether a person can consistently turn those capabilities into outcomes that others cannot easily reproduce or verify.
Specialised learning therefore acquires a different economic role. Traditionally, it often taught people how to perform the work themselves. In an AI-dominated economy, some of its greatest value may lie in teaching people how to supervise work performed by intelligent systems.
The expert learns to recognise the normal case, diagnose the abnormal case, and determine the appropriate response. They become the pilot of an increasingly autonomous operation.
The moat is not absolute. AI may eventually acquire many of the same diagnostic capabilities, and some fields will require less human supervision than others. Nevertheless, while a meaningful gap exists between producing an answer and reliably evaluating it, specialised judgment can create an advantage.
6. The K-Shaped AI Economy: Who Rises and Who Falls?
The Pilot-Mode Specialisation Moat suggests a possible explanation for why AI could produce sharply diverging economic outcomes.
Consider two workers whose jobs involve similar amounts of intellectual activity. One performs tasks that AI can reproduce with increasing reliability. The other uses AI to expand the reach of specialised expertise, directing the system and evaluating its results. As the technology improves, the first worker may face greater competition from automated output, while the second may be able to serve more clients or manage larger projects.
The difference is not simply that one worker uses AI and the other does not. Both might use it extensively. The difference lies in their economic relationship with the technology.
The first competes with the output AI can produce. The second uses AI to multiply their ability to produce and validate valuable outcomes.
This is a plausible mechanism for a K-shaped recovery: the same technological advance creates opportunities for one group while weakening the bargaining position of another. Experts who can direct AI effectively may become more productive, while workers whose contributions are easily reproduced may find their skills less scarce.
But the outcome is not predetermined. A domain expert may be displaced if AI can perform the relevant work end to end. A routine worker may move into a more valuable role by acquiring new skills. And even when AI increases a specialist’s productivity, the gains may flow to an employer, platform owner, or customer rather than to the individual who operates the system.
The central question is therefore not simply who benefits from AI. It is who captures the value created by AI, and what capabilities allow them to capture it.
7. The Moat Is Moving: When AI Learns to Handle Exceptions
The strongest objection to this argument is also the most important one: why assume that AI will remain unable to handle the exceptional cases that justify human expertise today?
AI systems are already capable of diagnosing certain errors, reviewing their own outputs, generating alternative solutions, and helping investigate unfamiliar problems. As their capabilities improve, the range of situations requiring human intervention may shrink.
If AI eventually performs routine operations, identifies failures, selects corrective actions, and verifies its results with sufficient reliability, the pilot-mode moat will narrow. The human’s role could become less necessary in domains where errors are detectable, consequences are manageable, and automated safeguards are effective.
This possibility does not invalidate the argument. It changes its time horizon.
The moat should not be understood as a permanent boundary between what humans can do and what machines cannot. It is a moving boundary that depends on the current capabilities of AI, the reliability demanded by the task, and the consequences of failure.
As AI improves, yesterday’s specialised knowledge may become tomorrow’s automated function. Professionals must therefore continue learning, not simply to maintain a fixed advantage, but to adapt as the source of scarcity changes.
The most durable advantage may come from combining domain knowledge with the capacity to learn, verify, adapt, and take responsibility in environments where the cost of error remains significant. Even these capabilities may eventually be partly automated, so no individual skill should be assumed to be permanently protected.
8. The New Physics of an AI-Dominated Economy
The phrase new physics describes a change in the relationships governing economic value. It is a metaphor, not a claim that economics has acquired new physical laws.
When execution is scarce, the ability to perform a task commands value. When execution becomes abundant, that relationship weakens. The value of expertise may increasingly depend on the quality of the decisions it informs and the amount of productive capacity those decisions direct.
A useful conceptual relationship is:
Economic leverage ≈ Value of outcomes influenced ÷ Human effort required
This is not a measured law or a complete economic model. It captures a simple possibility: a person may influence a large volume of valuable work without personally performing most of the operations.
A specialist who directs a capable AI system might accomplish in hours what previously required a team working for weeks. If the specialist’s judgment improves the reliability and value of that output, their productive leverage increases.
But the relationship has another side. If AI can make the same decisions without the specialist, their individual contribution may lose value. Greater productivity does not automatically produce greater wages, profits, or bargaining power. The distribution of economic gains depends on ownership, competition, market demand, and the ability of workers to capture the benefits they create.
The new scarcity is therefore not simply human intelligence or AI capability. It is the combination of valuable objectives, reliable outcomes, effective control, and the economic power to capture the resulting value. As technology advances, the relative scarcity of each component will change.
9. From Executor to Navigator: The Future of Specialised Learning
If this analysis is correct, specialised learning will need to evolve alongside AI.
Learning every procedure may become less necessary when AI can explain and execute many procedures on demand. But professionals still need enough understanding to judge whether the procedure is appropriate, recognise when the output is unreliable, and decide which evidence or expert review is required.
The emphasis shifts from memorising every operation toward understanding systems, constraints, failure modes, and decision points. The aim is not to abandon foundational knowledge. Without it, a person may be unable to recognise an error in the first place. Rather, foundational knowledge becomes the basis for directing and evaluating increasingly capable tools.
This shift will not affect every profession equally. In some fields, practical execution is itself essential to developing judgment. In others, simulation, AI-assisted practice, or supervised analysis may allow people to acquire useful expertise more quickly. High-stakes professions will still need appropriate training, experience, validation, and accountability; access to an intelligent assistant cannot substitute automatically for professional competence.
The challenge is to learn what remains necessary as AI takes over more of what was previously taught. That means developing the ability to navigate intelligent systems while remaining prepared for the moments when their outputs cannot safely be accepted at face value.
10. The New Expert: From Executor to Navigator
The surgeon, the pilot, and the AI-enabled specialist illustrate three aspects of the same transformation.
The surgeon demonstrates why knowing a procedure is different from knowing how to handle complications. The pilot demonstrates how a human can direct a highly automated system without manually performing every operation. The checkpoint principle identifies where human expertise can retain disproportionate importance: at the decisions that determine whether the larger process succeeds.
Together, these ideas suggest a different way to understand the future of professional expertise. The most valuable individual may not be the one who personally performs the greatest number of tasks, but the one who can direct intelligent systems toward reliable and valuable outcomes.
This does not mean that every expert will prosper, that routine workers have no future, or that human judgment will always remain superior to AI. It means that as execution becomes easier to obtain, the economic value of specialised learning may increasingly depend on the decisions it enables a person to make.
The Pilot-Mode Specialisation Moat is therefore a moving advantage, not a guarantee of permanent protection. It exists wherever specialised knowledge enables a person to direct, evaluate, and correct AI more effectively than would otherwise be possible. As AI acquires those capabilities itself, the moat must move.
The central challenge for individuals, businesses, and education systems is to understand that movement before yesterday’s advantage becomes tomorrow’s commodity.
AI may automate the crew. The specialist’s moat lies in knowing which checkpoints matter, what to verify, and when to intervene.
And the ultimate question remains: if AI eventually learns to handle every important checkpoint as reliably as a human expert, what will become the next scarce capability in the economy?
Continue the Series
This article is Part 2 of The Future of Expertise in the Age of AI. Return to Part 1: When Knowledge Stops Being a Moat: The Future of Careers in the Age of AI for the opening argument about the changing economic value of specialised knowledge.