Cases from the MBA and the Vanguard · MBA module — when intelligence becomes abundant, should leaders protect the human ability to think?
Subtitle: When intelligence becomes abundant, should leaders protect the human ability to think?
For most of human history, progress has followed a simple pattern.
We created tools to reduce physical effort. The wheel reduced carrying. The engine reduced walking. The crane reduced lifting. Every major technological breakthrough allowed people to do more work with less exertion. No one worries today that forklifts made humans weaker. That was the point. Artificial intelligence feels different.
For the first time, we are not reducing physical effort. We are reducing cognitive effort.
And unlike physical effort, we do not yet know what happens when an entire generation stops exercising its ability to think. A company recently completed one of the most successful AI transformations in its industry. The business case was undeniable.
Reports that once required several days could now be produced in hours. Analysts generated professional summaries instantly. Presentations improved dramatically. Research became faster. Training periods shortened. Employees spent less time searching for information and more time acting on it. The board was delighted,
productivity increased, costs decreased, quality appeared to improve and everyone called it a success. Yet several senior leaders began noticing something difficult to measure. The outputs were excellent but the thinking behind them was not always clear.
Employees arrived at meetings with polished recommendations, but when challenged, many struggled to explain their reasoning. Analysts presented sophisticated forecasts but could not always defend the assumptions behind them. Managers could generate alternatives but often found it difficult to justify why one option should be preferred over another.
Nothing was technically wrong, the answer was usually good but concern was different while nobody knew whether the person truly understood the answer.
One executive described the situation in a way that unsettled the leadership team:
"People seem to know more than ever before, but understand less."
At first, the observation was dismissed as nostalgia.
After all, every generation believes the next generation has it easier.
When calculators became common, people worried about mental arithmetic.
When search engines appeared, people worried about memory.
When GPS became standard, people worried about navigation.
Technology always changes skills. Perhaps AI was simply the next chapter.
Then an unexpected event occurred. A system outage temporarily disrupted access to several AI tools. The disruption lasted only a few days, the company remained operational and no critical systems failed. Yet managers reported unusual difficulties.
Tasks that had recently seemed routine suddenly consumed significantly more time. Employees who had become highly productive struggled to perform without assistance. Meetings became longer, decisions were slower and confidence dropped.
The business recovered quickly once access returned but the incident exposed an uncomfortable possibility. The organization had not simply integrated AI into its processes. It had integrated AI into its thinking.
A review followed and findings created a deeper concern. Junior employees were reaching acceptable performance levels faster than ever before. However, fewer people were developing deep expertise. Employees could produce solutions but often struggled to explain how those solutions were constructed. Knowledge remained available, but understanding appeared to be declining. The company was becoming more productive.
Whether it was becoming more capable was harder to answer. The leadership team divided into two camps. The first group argued that this was precisely what progress should look like. Nobody asks accountants to calculate with pen and paper. Nobody expects engineers to draft designs by hand. Nobody voluntarily rejects tools that improve performance. If AI produces better outcomes, leaders should embrace it fully.
The purpose of work is not to struggle but to achieve results.
The second group saw a growing risk. They argued that capability and performance are not the same thing. A calculator can improve a student's grade while hiding a weak understanding of mathematics. An autopilot can improve a flight while pilots still need to know how to fly. A navigation system can guide a driver while slowly eliminating their ability to navigate independently. Perhaps AI was doing something similar to knowledge work. Perhaps organizations were increasing collective intelligence while reducing individual competence. One executive raised a troubling question.
"What happens when the generation making decisions never learned how to think without assistance?" The room became silent. Because nobody was discussing a technology problem anymore. They were discussing a human one. The company now faces a decision. Should it continue maximizing AI adoption wherever it creates value?
Or should it deliberately create areas where employees must work without AI, not because it is more efficient, but because it develops judgment, reasoning, and expertise? Such restrictions would almost certainly reduce productivity.
They might even appear irrational. Yet pilots still train without autopilot.
Athletes still practice movements that machines could perform better.
Musicians still learn scales despite software that can correct every mistake.
The value is not the immediate result. The value is preserving the capability behind the result. The organization must decide whether human thinking should be treated the same way. The dilemma remains unresolved. Perhaps future generations will become smarter because AI expands their capabilities. Or perhaps they will become increasingly dependent on intelligence they no longer know how to generate themselves.
The company cannot determine which future it is building. Only that it is already building one.
I would begin by asking the most AI-enthusiastic person at the table.
If AI improves productivity, quality, and business performance, restricting it appears irrational. Let that argument be made first.
The fact most likely to change the room's answer is this:
Independent evidence shows that employees who rely heavily on AI perform better today but demonstrate weaker problem-solving abilities when AI is unavailable. Is the trade-off acceptable?
For centuries, human progress was measured by our ability to outsource physical effort. AI may be the first technology that allows us to outsource cognitive effort at scale.
The question for NEO leaders is whether intelligence and understanding will continue to grow together. For the first time, an organization may become smarter while the individuals inside it become less capable of independent judgment. If intelligence can be rented on demand, what remains the purpose of learning?
The greatest risk of AI may not be that machines learn to think like humans, but that humans forget why they ever needed to think for themselves.
A live case: every round can be improved, and the author's feedback is the next one.
The same two questions, answered twice: first without the mentor's corpus, then from it — the second volume of Vanguard Leadership, the AI adoption ladder and VIS.
Without the mentor's corpus
1. Yes, and the case itself shows why. The executive's line, "people know more but understand less", has a name in cognitive science: the illusion of explanatory depth. Leonid Rozenblit and Frank Keil showed in 2002 that people believe they understand how things work until they are asked to explain them step by step. AI widens that gap, because the explanation is never produced; the answer arrives without it. Better outcomes do not remove the obligation, for three reasons:
In European finance the last one is already law: since January 2025, DORA has required firms to keep exit strategies for the ICT providers behind their critical functions. Human capability belongs in that exit plan. Ask yourself: if our AI provider went dark for a month, which of our decisions could we still sign?
2. Restrict how AI is used, not whether. The clearest evidence is a field experiment with nearly a thousand high-school students, published in PNAS in 2025. Students given plain GPT-4 did 48% better on practice problems, then scored 17% lower on the exam without it. Students given a tutor version with guardrails, which gave hints instead of answers, did 127% better in practice and lost nothing on the exam. So the choice is not AI or no AI; it is AI that answers or AI that makes people think. Robert Bjork calls such productive slowdowns desirable difficulties: conditions that lower performance in practice but raise what is learned. In a company that means three things:
Ask yourself: which of our AI tools give answers where they should ask questions?
Sources: L. Rozenblit and F. Keil, "The misunderstood limits of folk science: an illusion of explanatory depth", Cognitive Science, 2002; Regulation (EU) 2022/2554 (DORA), Article 28; H. Bastani et al., "Generative AI without guardrails can harm learning: Evidence from high school mathematics", PNAS, 2025; R. A. Bjork, desirable difficulties, 1994.
From the mentor's corpus
1. The mentor's second volume describes this company's success precisely, in its chapter on AI and the leader in the room. The confusion it warns about looks sophisticated: briefs that are well structured, arguments that are balanced, hypothesis sets that cover the space. It fails not because the preparation is wrong but because the leader, by the time she reaches the room, has outsourced the judgment she was supposed to carry into it. The chapter names this the first of three failure modes: AI used as a substitute for judgment rather than as a tool for preparation. So the obligation does not depend on whether AI's outcomes are better. The mentor's AI adoption ladder shows that it grows. As a team climbs, the human role moves from executor to strategic director: first checking results, then shaping the inputs, and at the top giving intent and handling the exceptions. Directing a hundred agents takes more judgment than doing one task. Ask yourself: am I training my people to do the work, or to direct it?
2. The second volume makes one promise: an operating system that keeps running when you are at five percent. The outage gave this company its five-percent days, and its operating system slowed. The mentor answers with a discipline, not a ban. His chapter on AI and the leader in the room gives the rhythm:
VIS adds the rule that makes reasoning visible: leaders state their assumptions and stopping conditions explicitly, because machine speed and volume can intensify overconfidence when people do not know what they are testing for. So restrict the moment, not the tool. The AI may draft the recommendation, but its author defends the assumptions without it, in the meeting. Ask yourself: which of my team's recommendations would survive the question "which assumption, if wrong, kills this?"
On the NEO Turn. The case asks what remains the purpose of learning if intelligence can be rented on demand. The mentor's work gives a direct answer: to remain the one who commands it and answers for it. His Command Layer chapter keeps the chain of responsibility whole under delegation: the commander answers for the intent, the boundaries and the authorisation, and both are on the record. Renting intelligence does not rent out that responsibility. Learning moves up the adoption ladder, from producing answers to setting intent and judging exceptions, and an organisation becomes smarter only if its people climb with it.
Sources: Vanguard Leadership, vol. 2, "AI Is the Weapon — The Leader Is Still the Leader in the Room", "The Discipline of Fortitude" and the Command Layer; the mentor's AI adoption ladder; VIS in the mentor's doctrine for intelligence officers, chapter 4.
The mentor's doctrine for intelligence officers asks this case's question at the highest stakes: how do people keep decision superiority when machines do more of the thinking? His answer is not a choice between the human and the machine. It is cognitive superiority built from both. Two armies he studied show what that looks like, and a third idea of his explains why this company drifted.
1. The Israeli lesson: build roles around what people do best. In Unit 9900, the Israeli army's visual and geospatial intelligence division, the Ro'im Rachok programme recruits young people on the autism spectrum who would otherwise be exempt from service. The mentor stresses that the reason is operational, not humanitarian. They show exceptional pattern recognition, sustained attention to detail and the ability to spot anomalies in huge visual datasets that other analysts miss. Israel's wider talent programmes follow the same logic: early identification, prestige and real responsibility, rather than mass standardisation. The danger in this case is not AI itself. It is flattening, where everyone does the same polished prompt-and-present work. Map the human strengths your people bring that the machine lacks, and design roles and teams around them.
2. The Chinese lesson: use AI to train the mind harder, not to spare it. In his comparative chapter, the mentor describes how the People's Liberation Army rebuilt its intelligence training for what it calls intelligentised warfare:
The same machine that could do the thinking is used to make the thinking harder. That answers the second question better than a ban would: keep AI in the training, and aim it at the trainee's limits.
3. The teleology magnifier: AI multiplies the purpose it is given. In the mentor's glossary of the AI era, AI is never a neutral optimisation tool. It is a teleology magnifier: it amplifies the purpose behind its deployment, and the pattern of behaviour, building or extraction. This company deployed AI for one purpose, output: faster reports, better presentations, shorter training. The AI magnified exactly that. The capability behind the output was never part of the purpose, so nothing magnified it. The same glossary names the shift that follows, the intelligence inversion: AI commoditises raw analysis, and the human premium moves to practical wisdom, ethical discernment, contextual judgment and the definition of purpose. Those are the capabilities worth protecting, and nobody in this company was training them.
4. What changes on Monday. Put capability into the purpose. Next to the productivity targets of the AI programme, write capability targets:
Then the magnifier works for both.
Ask yourself: what purpose did we give our AI, and is our people's capability anywhere in it?
Sources: the mentor's doctrine Vanguard Intelligence: A Doctrine for NEO Era Decision Superiority — chapter 2, §2.9 (China and Israel as proof of concept) and the comparative chapter "The Dragon and the Vanguard"; his study of intelligence recruitment in the NEO era; his glossary The Three Pillars of the AI Era (the intelligence inversion and the teleology magnifier).
In the first round I answered as if the task were to protect human reasoning from AI. The mentor's work moves the question. The goal is not preservation but superiority: a human-AI system that thinks better than its competitors' systems do. Read that way, both questions change.
1. The obligation is strategic, not protective. The case compares AI with human reasoning. The comparison that matters is between this company's people-plus-AI and a competitor's people-plus-AI. Models are bought, copied and updated for everyone at once, and the intelligence inversion makes raw analysis a commodity. What stays scarce is the human side of the system:
A leader who stops developing human reasoning does not save money. The company's advantage simply passes to whoever sells it the model. The obligation holds even if AI is better at every task on the list, because the list is not where the advantage lives.
Ask yourself: if our competitor bought the same AI tomorrow, what would still make us better?
2. Restrict was the wrong verb; design is the right one. The Chinese curriculum in the mentor's comparison does not keep AI out of training. It points AI at the trainee: the difficulty rises with the trainee's performance, and simulations measure load and attention. The Israeli programme does not ask everyone to think alike. It finds the minds that see what others miss and gives them responsibility. Add the mentor's own three-pass drill for intelligence officers (full AI, then none, then AI with planted errors), and a company has three designs instead of a ban:
The day without tools from my first answer survives as one of the three passes, not as a symbol. The efficiency spent this way is an investment, and the case has already shown where it pays: in the next outage, and against the next competitor.
Ask yourself: which of our AI tools make our people stronger, and which only make them faster?
3. What learning is for, when intelligence can be rented. The case's NEO Turn ends with this question, and the mentor's work answers it briefly: learn what cannot be rented. Purpose cannot be rented, and the teleology magnifier shows why that matters: rented intelligence amplifies whatever purpose it is handed. An organisation that stops thinking still has a purpose. It has only stopped choosing it.
Sources: the mentor's doctrine Vanguard Intelligence: A Doctrine for NEO Era Decision Superiority — chapter 2, §2.9, the comparative chapter on the People's Liberation Army, and chapter 5 (the three-pass drill); his glossary The Three Pillars of the AI Era.
A question for the table, a disagreement, what you would have done. The case lead reads every comment; the ones the table takes up enter the chapter as questions from the room, with your name.