Cases from the MBA and the Vanguard · MBA module — when AI makes one leader look like ten
Subtitle: When AI Makes One Leader Look Like Ten
The first sign that something had changed was not the quality of the work.
It was the speed.
Within three months, one middle manager had become impossible to keep up with. He answered emails within minutes, produced meeting summaries before anyone else had finished processing the meeting, and delivered polished presentations overnight. When senior leadership asked for a market analysis on Thursday afternoon, he had a twenty-slide version ready Friday morning.
His team initially loved it.
He was removing friction. He was making decisions faster. He was protecting them from administrative work that had consumed much of their time.
Then people realized how he was doing it.
Six months earlier, the company had introduced an internal AI assistant. Employees were encouraged to experiment with it, but there were few formal rules beyond confidentiality. The manager had gone further than anyone else. He built a workflow around the system: meeting transcripts went in, customer feedback went in, reports went in, and structured decision documents came out.
His results improved.
His department became one of the company's most productive. Client satisfaction increased. Deadlines were met more consistently.
The CEO noticed.
"We need the other managers to learn what he's doing," she said at the quarterly leadership meeting.
The HR director disagreed.
"We need to understand what he's doing first."
The problem became visible when two of his recommendations contained subtle errors. One market statistic was outdated. One customer assumption had been inferred from incomplete information. Neither caused serious damage, and both were corrected.
When challenged, the manager was straightforward.
"I use AI because it makes me faster. But I'm still accountable for the decision."
The CEO asked him how much of the work was actually his.
"I use AI to get to the first 80 percent," he said. "The last 20 percent is where I add the judgment."
"What exactly is the 20 percent?"
He smiled.
"That's the part you can't automate."
The answer sounded reassuring.
It also exposed the real problem.
The company had always known who was accountable for a decision. It had never needed to know how much of the thinking behind it came from a person, a team, a consultant, a spreadsheet — or now, a machine.
Meanwhile, other managers had started copying him. Some were quietly using public AI tools with sensitive information. Others were producing impressive documents they could not fully explain. A few had begun skipping slower parts of management — mentoring, discussion and disagreement — because AI could produce an answer before the team had finished defining the question.
One senior manager put the concern bluntly:
"We are rewarding the person who can produce the most convincing answer. What happens when the scarce capability is knowing which questions deserve to be asked?"
The CEO now faced a decision.
She could formalize the manager's workflow and make it the new standard.
She could slow AI adoption until governance caught up.
Or she could allow experimentation while changing how leadership performance was measured.
Each option carried a cost.
Stopping the manager could discourage exactly the experimentation the company needed. Scaling his approach without understanding it could multiply hidden risks just as quickly as it multiplied productivity.
The CEO postponed the decision for one week.
She asked the manager to document his workflow, HR to propose an AI governance framework, and the CFO to answer one uncomfortable question:
If AI makes one manager five times more productive, what exactly are we paying the other managers to do?
Open with: "If this manager had become five times more productive by hiring five junior analysts, would we have the same concern?"
Ask the strongest AI advocate to speak first, then the strongest governance voice.
The fact that should change the room's answer is whether the manager can explain and defend his decisions without the AI present.
The next version of this problem may not involve a manager using AI as a tool. An AI agent may monitor performance, prepare decisions, communicate with customers and initiate actions itself.
Then the question changes from "Who used AI?" to "Who designed the system, who could stop it, and who remains responsible?"
When intelligence becomes abundant, the scarce resource may no longer be the ability to produce an answer — but the judgment to know which answer deserves to become a decision.
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 AI adoption ladder, the first volume, the agent charter, the decision doctrine and the Diagnostic Layer.
Without the mentor's corpus
1. Reward what you would reward if he had hired five analysts: the quality of his decisions and what his leverage does for others, not his speed. Steven Kerr named the trap in 1975, in "On the Folly of Rewarding A, While Hoping for B". This company rewards the most convincing answer while hoping for the best questions, and one of its senior managers has already seen it. Research on AI at work points to where the value lies. When a large customer-support operation gave its agents an AI assistant, productivity rose 14% on average and 34% for novices, with little gain for the best people, because the tool spread the best agents' practice to everyone else. So the manager's real contribution is not being ten people. It is a workflow that could make the other managers better. Pay him for turning it into a shared, governed asset, and judge everyone by decisions that hold up. That also answers the CFO: the other managers are paid to choose the questions, develop their people and own their decisions. Ask yourself: am I rewarding the answer, or the judgment about which answer should become a decision?
2. He owes the organisation the provenance of a decision, not the diary of how he made it. The clearest evidence for why comes from a field experiment with 758 BCG consultants. On tasks inside what the authors call AI's jagged frontier, consultants with AI finished 12.2% more tasks, 25.1% faster and at over 40% higher quality. On a task just outside it, they were 19 percentage points less likely to get it right. His two errors sit exactly on that edge: a statistic past its date, and a customer assumption inferred from incomplete data. So the transparency owed is narrow and strict:
Beyond that, his prompts, drafts and workflow are his craft. Accountability becomes control when it asks for the drafts instead of the sources, or approves every step instead of checking a sample. Ask yourself: if I read one of my team's decisions a year from now, could I tell which parts were facts, which were guesses, and which were the machine's?
Sources: S. Kerr, "On the Folly of Rewarding A, While Hoping for B", Academy of Management Journal, 1975; E. Brynjolfsson, D. Li and L. Raymond, "Generative AI at Work", Quarterly Journal of Economics, 2025; F. Dell'Acqua et al., "Navigating the Jagged Technological Frontier", 2023.
From the mentor's corpus
1. The mentor's AI adoption ladder describes this manager precisely, and predicts his errors. He works at its second level, where AI produces finished deliverables in parallel and the person reviews them at the end. The ladder names that level's bottleneck: validation becomes the major time sink, and two subtle errors are what a validation bottleneck looks like. The next level changes the job. The focus moves from checking results to improving inputs, and the level requires institutional governance written into specifications the AI can read. That is what HR has been asked to draft, and it is what the company should reward: not one person's speed but the team's climb. The ladder's own measure is how many agents a team, not a star, can command well. The first volume adds the human measure: how many people can execute on your intent without direct oversight? A leader who speeds up while mentoring and disagreement disappear around him scores low on both. Ask yourself: would my team be stronger or weaker if my AI workflow disappeared tomorrow?
2. Two rules from the mentor's work fix both errors without policing anyone:
So the transparency owed is the provenance and the status of what a decision rests on, not the prompts behind it. The mentor's Diagnostic Layer shows where accountability would tip into control. It measures something other than outputs: whether the process that produced them was trustworthy. In practice that means a quarterly sample-check of decision documents against their sources and statuses, not an approval of every step. The rest is his craft. Ask yourself: which of my decisions would pass an audit of its sources today?
On the NEO Turn. The case ends with three questions for the day an agent acts on its own: who designed the system, who could stop it, and who remains responsible? The mentor's agent charter answers each on one page before the agent runs:
The manager in this case is the natural author of the first charter in his company. Asking him to write it would test the twenty percent he says cannot be automated.
Sources: the mentor's AI adoption ladder; Vanguard Leadership, vol. 1, §2.4; the Vanguard Task-to-Agent Mapping Protocol (the agent charter, the VIS provenance rule); the mentor's decision doctrine of 22 August 2026; Vanguard Leadership, vol. 2, the Diagnostic Layer.
While one manager works alone and waits for 100 percent certainty, another, with AI and 70 percent, can do ten times more work. That is the point of this case. In the NEO era, being 80 percent sure of a decision is more than enough, and 70 percent is fine. The rule is commonly attributed to Jeff Bezos, though it is older: if you have about 70 percent of the information you wish you had, act. Waiting for 90 percent or more makes you too slow, and in business being slow is expensive for sure.
What makes it safe is not certainty but iteration: fast feedback, constant adjustment and adaptation, the OODA loop on repeat. Most decisions can be reversed or adjusted. The discipline is to know which ones cannot, and to slow down only for those.
The mentor's point is arithmetic before it is attitude, and it is stronger than it sounds.
Two multipliers, not one. AI multiplies the work behind each decision: the analysis that took a week takes an evening. The 70-percent rule multiplies the number of decisions. The last 30 percent of certainty is usually the most expensive part to buy, and the fast manager simply does not buy it. One multiplier alone gives a faster manager; both together give the order of magnitude the mentor names.
Speed buys accuracy, over time. John Boyd's point, on which the mentor's doctrine builds, is that victory goes to whoever cycles through observing, orienting, deciding and acting faster than the opponent. Every cycle is also a lesson. Put numbers on one quarter, as an illustration:
He ends the quarter with seven decisions that held, three lessons, and a better model of his market than the one who waited. By the eleventh decision he is more likely to be right, not less. Certainty is not what the slow manager gets for waiting; it is what the fast one earns by iterating.
What this changes in the case. The risk this company should fear is not its fastest manager. It is the manager who still waits for certainty while a competitor's managers decide at 70 percent with AI. So the CFO's question has an honest answer: the other managers are paid to become the same kind of manager, and the company's job is to make that safe for all of them. The two errors show the condition. The tenfold advantage holds only while mistakes stay small and cheap to correct. That is why the dated source and the marked assumption from my first round matter: they turn a correction into a week's work rather than a year's discovery. And it is why the mentor keeps one exception. Where a decision cannot be reversed, as when confidential data goes into a public tool, 70 percent is not enough.
Ask yourself: in the time I spend waiting for certainty on one decision, how many could a manager with AI make, correct and learn from?
Sources: the mentor's round above; J. Boyd's OODA loop, as the mentor's article on NEO Cotruglian philosophy for the International Leadership Journal reads it; J. Bezos, letter to Amazon shareholders, 2016 (the seventy-percent rule).
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.