Why Executives Need AI Literacy

Before reading on, make a guess. Think of the knowledge workers in your organization. What share of them, would you say, already use generative AI for at least 30 percent of their daily work?

In late 2024 McKinsey asked that question twice. It asked 238 C-level executives to estimate the share, and it asked 3,613 employees what they actually did. The executives estimated 4 percent. The employees reported 13 percent, more than three times as many¹. The two numbers measure different things: the first is leaders guessing about other people, the second is employees reporting on themselves.

If your guess was low, you are in good company, and that is the problem. The people who set the budget, the risk appetite and the strategy had the least accurate picture of how the work beneath them was already changing. McKinsey’s own conclusion was blunt: employees were ready, and the biggest barrier to scaling was leadership¹. Read that with some care. The data come from one online survey, 81 percent of it in the United States, run by a firm that sells AI transformation work, and a finding that leadership is the bottleneck suits that business. Treat 4 against 13 percent as indicative of a direction rather than a precise measure; the direction is what matters here.

The previous chapter asked why a program like this deserves an executive’s time when specialists understand AI far better than any leader needs to. This chapter answers it. Specialists can tell you what AI can do. They cannot decide what your organization should do with it, and they will not be the ones held accountable when the decision goes wrong.

The decisions have left the technology department

For most of the history of corporate computing, leaders could delegate technology with a clear conscience. Few chief executives needed to understand how a database stores its records. They needed to know what it cost, whether it worked and who was responsible for it, and the specialists handled the rest.

AI does not stay in that box. It drafts what customers read, shapes what employees do all day, touches confidential data and makes or informs decisions that used to belong to people. The consequences of an AI choice therefore land in places that no technology team owns: the product, the brand, the workforce, the risk register and the competitive position. The choices that matter most are business choices with a technical component, not the other way round.

Each of these decisions arrives at the executive table whether or not the people around it understand AI. A proposal for an AI platform is an investment decision. A request to let staff use a public chatbot is a risk decision. A plan to automate part of a function is a people decision. The decision will be made either way. Literacy decides whether it is made well.

What literacy means for a leader

AI literacy is easily misheard as a technical skill, so it helps to be exact. The researchers Duri Long and Brian Magerko, in a widely cited definition, describe it as a set of competencies that let a person critically evaluate AI technologies, communicate and collaborate effectively with AI, and use it as a tool². Notice what is missing from that list: building models, writing code and memorizing architectures.

For an executive, those competencies reduce to three capacities. Each depends on the one before it.

Judgment is a working sense of what today’s AI does well and where it fails. Its most important lesson is that capability and reliability are different things. A system can produce a fluent, confident and plausible answer that is simply wrong. As AI Is Changing Everything showed, AI’s competence is jagged: excellent at some tasks and poor at neighboring ones that look just as hard. A leader who assumes that a system that writes well must also know what it is writing about will approve the wrong things.

Questions are where judgment becomes useful. The literate executive does not know more answers than the specialists in the room. The difference is the questions: the ones that turn an impressive demonstration into a decision that can be defended.

Accountability is the part that cannot be handed to anyone else, and the reason the first two matter.

Better questions, not more answers

The fastest way to see literacy at work is to compare the questions that typically reach an AI proposal with the questions a literate leader asks instead.

None of the questions on the right requires technical training to ask. Each requires knowing that the answer matters. “What happens when it is wrong?” only occurs to someone who knows that it will sometimes be wrong. “Including the people who check it” only occurs to someone who knows that AI output often needs checking. That is the practical shape of literacy: a short list of questions that a leader knows to ask, and knows how to judge the answers to. Later modules supply the detail behind each one, from evaluation to economics.

The cost of not having it

AI Is Changing Everything described the two stances that waste the AI years, complacency and panic. Low literacy is the usual reason leaders fall into one or the other, and it adds a third failure that is easier to miss.

Abdication looks like caution. Leadership calls AI a technology matter and waits. But the work does not wait, as the gap between 4 and 13 percent shows. Employees adopt tools on their own, and the important choices, which data goes where and which outputs are trusted, get made one person at a time, without anyone deciding.

Overreach looks like ambition. Leadership declares that AI must be everywhere, and every function launches its own pilots. The pattern is now measurable. In S&P Global Market Intelligence’s 2025 survey of more than 1,000 organizations in North America and Europe, 42 percent had abandoned the majority of their AI initiatives that year, up from 17 percent in 2024. The average organization scrapped 46 percent of its proofs of concept before production³. The first figure counts companies that gave up on most of their initiatives; the second is the share of early prototypes a typical company dropped. Some of that is healthy experimentation. Much of it is money spent before anyone asked what the initiative was for.

Over-trust looks like confidence, and it is arguably the most dangerous, because it feels like progress. A leader who has seen AI produce brilliant work comes to believe its output, and approves it into places where a wrong answer is expensive. The story later in this chapter shows how quickly that cost can arrive.

Accountability does not delegate

A common objection to executive AI literacy is that the organization already has a chief technology officer, a data team and consultants. It does, and they matter. They can design, build, test and run AI systems. What they cannot do is take over the leadership team’s accountability for the decision to use them.

Two leading management standards say so directly. The US National Institute of Standards and Technology’s AI Risk Management Framework, a widely used voluntary framework, lists among its governance outcomes that executive leadership takes responsibility for decisions about the risks of developing and deploying AI⁴. ISO/IEC 42001, the international standard for AI management systems, requires top management to demonstrate leadership and commitment to the AI management system⁵. Neither asks executives to become engineers. Both assume that executives understand enough to own the decisions.

In practice, the line falls between how and whether. How a system is built, tested and run can be delegated. Whether to use AI for a purpose, why, how much risk is acceptable and who answers for the result cannot. AI Decision Rights and Accountability, in Module 04, works through where that line sits in an organization. The point here is simpler: an executive who cannot follow the argument for an AI decision cannot really own it, and an executive who cannot own it should not sign it.

The board’s share

The same logic reaches the boardroom, where literacy is often thin. In a 2026 study by The Conference Board, only 23 percent of executives described their board as highly fluent in AI, and disclosed AI expertise among S&P 500 directors stood at 2.7 percent in 2025⁶. The first figure is how executives judge their board; the second counts directors whose AI expertise their company discloses.

Directors do not need to be AI experts, and boards rarely should be. But oversight has a legal edge. In the United States, the Delaware Caremark doctrine holds that directors can breach their duty if they make no good-faith effort to put a reporting system in place for mission-critical compliance risks, or ignore the red flags it raises. In Marchand v. Barnhill (2019), the Delaware Supreme Court let such a claim proceed against the directors of an ice cream maker whose board had no regular reporting on food safety, although food safety was central to the business⁷. Where AI becomes mission-critical, a board that cannot ask about AI risk, or would not understand the answer, is exposed. AI Governance Committee and AI Governance Office, in Module 07, sets out the reporting that boards should expect.

Literacy is also not the same at every level. The foundation, the vocabulary and the core judgments should be shared across the organization. The depth changes with the role.

Story: a fluent answer at launch

This story is real, and it is told as a decision. Put yourself in the leadership of a technology company in February 2023.

A rival’s chatbot has become the most talked-about product in the industry, and another rival is about to put a chatbot into its search engine. Your company has one of the deepest AI research teams in the world and a capable conversational model of its own. The plan is to announce it now, with a short demonstration of the model answering a question, and to open it first to a group of outside testers before a wider launch. The demonstration is ready.

You have two options. Announce this week, with the demo, and show the market you are not behind. Or hold the demo until every answer it shows has been checked by someone outside the team that built it, and accept a week of headlines saying you are slow.

On 6 February 2023 Google announced Bard, “an experimental conversational AI service”, and said it would open first to trusted testers⁸. The promotional demonstration showed Bard answering a question about new discoveries from the James Webb Space Telescope that a parent could share with a nine-year-old. One of its answers said the telescope had taken the very first pictures of a planet outside our solar system. It had not. That image was taken in 2004 by the European Southern Observatory’s Very Large Telescope⁹.

Reuters reported the error on 8 February. Alphabet’s shares fell 7.7 percent that day, about 100 billion dollars of market value. A company event the same day also failed to impress investors, so the error was not the only cause, but it was the one that defined the coverage⁹. Google’s response was that the episode highlighted the importance of a rigorous testing process, which it was beginning that week with its trusted testers¹⁰.

The lesson is not that Google lacked expertise. Few organizations had more. It is that the failure happened at the point where expertise ends and leadership begins: the decision about what a fluent answer must pass through before it speaks in the company’s name, in public, under competitive pressure. That is a judgment about reliability, a question anyone in the room could have asked (“who checked these answers?”) and an accountability that sat with the people who approved the launch. All three are literacy, and none is technical.

What this means for leaders

Executive AI literacy is not a credential. It is a habit of mind applied to real decisions. Four practices follow from this chapter.

1. Close your own perception gap. Find out how AI is already being used in your part of the organization before you decide anything about it. Ask the people doing the work, not only the people reporting on it.
2. Bring the questions, not the answers. Take a short list of literate questions into every AI decision: the problem, the quality needed, what happens when it is wrong, the full cost and who owns the result.
3. Keep the “whether” with you. Delegate how systems are built and run. Keep whether, why and how much risk.
4. Make the board able to oversee. Ask what the board would need to see to know whether AI risk is under control, and make sure someone is building that report.

Check yourself

1. AI literacy for executives means learning to code or to build models.MythFact
2. Leaders tend to underestimate how much their employees already use generative AI.MythFact
3. Once a chief technology officer is in place, AI risk can be left to the technology team.MythFact
4. A fluent, confident AI answer can usually be treated as a reliable one.MythFact
5. To meet their oversight duty, directors must themselves be AI experts.MythFact
6. Under the amended EU AI Act, organizations must guarantee a specific level of AI literacy among their staff.MythFact

Reflection: your last AI decision

What comes next

Literacy lets a leader understand AI decisions. It does not yet make that leader the person the organization follows through them. Knowing the right questions is awareness; acting on the answers, consistently and in public, is leadership. The next chapter, From AI Awareness to AI Leadership, describes the behaviors that make that difference.