The Wrong Question

The meeting had ended, but the work had not. That scene remains the right place to begin because it contains the central tension of the AI era: a machine can make knowledge work look finished before the human work of meaning, consequence, and responsibility has even begun.

The argument of this expanded edition begins with a distinction that has become more important as AI systems have improved. Intelligence, as we once encountered it in professional life, was scarce partly because producing competent cognitive work required time, training, and access to people who possessed the necessary expertise. Generative AI changes that economics. It does not make wisdom abundant. It makes many forms of plausible cognitive production cheaper, faster, and easier to obtain.

That shift changes where value accumulates. When drafts are scarce, producing a draft is valuable. When ten drafts can be generated in seconds, the scarce act becomes deciding which one is true, useful, distinctive, responsible, and worth pursuing. When analysis is expensive, producing analysis carries value. When analysis is plentiful, framing the right question and knowing what evidence deserves trust become more important.

This is not an argument that machines are becoming human or that human beings possess some mystical reserve that technology can never reach. It is an argument about economics, work, and responsibility. Abundance in one layer of a system creates scarcity somewhere else. AI increases the supply of generated intelligence. That makes judgment, context, selection, trust, accountability, relationships, and purpose more consequential.

The expanded edition develops that claim at three levels. First is the individual: what should a person learn, retain, delegate, and strengthen? Second is the team: how should people and machines divide cognitive labor without weakening expertise or responsibility? Third is the institution: how should organizations redesign work, apprenticeship, management, governance, and measurement when intelligence is no longer the bottleneck it once was?

The goal is not to defend every task humans currently perform. Much work should be automated. Some work should disappear. The goal is to distinguish unnecessary friction from necessary thought, and borrowed capability from owned capability. The central question is not whether we can use AI. It is whether the way we use it leaves us more capable of doing work that matters.

The meeting had ended, but the work had not. For two hours, people had talked about money, alumni, donors, trustees, institutional ambition, old frustrations, new possibilities, and the delicate question underneath nearly every strategy meeting: what can this organization actually make happen? There was a transcript. There were notes. There would eventually be an agenda for the next meeting. None of those things was the strategy. I gave the conversation to AI. Within minutes, it could identify themes that would have taken a human assistant much longer to organize. It could pull out names, ideas, unresolved questions, possible funding concepts, stakeholder groups, and next steps. It could turn a messy conversation into something that looked remarkably coherent. And that was precisely where the danger began. The coherent version did not know which suggestion had changed the temperature in the room. It did not know which relationship was strong enough to act on and which was merely aspirational. It did not know that one person's hesitation carried more institutional weight than another person's enthusiasm. It did not know the history behind a sentence that looked ordinary on the page. I did. So the next instruction was not, “Make this better.” It was, in effect: Now let me tell you what actually happened. That exchange captures the argument of this book. Artificial intelligence can increasingly perform astonishing amounts of knowledge work. It can draft, summarize, compare, classify, code, analyze, simulate, translate, and generate possibilities at a speed that would have been absurd to imagine only a few years ago. But useful work has never consisted only of production. Someone still has to know what matters. Someone has to recognize what the machine misunderstood. Someone has to decide which possibility deserves resources, which claim deserves belief, which risk deserves attention, which relationship must be protected, and which recommendation should become action. Someone has to own what happens next. That someone is the subject of this book. The question is no longer whether AI can do parts of your job. It can. The better question is what becomes more valuable about you when it does.