From AI Awareness to AI Leadership

In August 2025, Sweden’s prime minister, Ulf Kristersson, told the business daily Dagens Industri that he uses AI chatbots “quite often” for a second opinion. He asked them questions such as what other countries had done, and whether the government should think the complete opposite. His press secretary added that nothing security-sensitive went into the tools; they were a sounding board¹.

By the standards of most AI advice to executives, this was model behavior. Leaders are told to stop delegating their curiosity, to use the tools themselves and to learn by doing. Kristersson had done exactly that. The reaction was fierce anyway. One of Sweden’s largest newspapers attacked him in an editorial, and Virginia Dignum, a professor of responsible AI at Umeå University, objected that voters had elected him, not a chatbot, and warned that leaning on AI for simple things invites overconfidence².

Both reactions were reasonable, and that is the contradiction worth sitting with. The prime minister was more aware of AI, and more practiced with it, than most leaders. What the public wanted to know was something his fluency could not answer: who decides, on what evidence, and who answers for the result. Those are leadership questions. Awareness, and even personal skill, leaves them open.

Awareness changes what you know; leadership changes what you do

Most executives now sit somewhere on a short ladder. At the first rung, they are aware: they know AI matters, their competitors are investing and their board is asking about it. At the second, they are literate: they understand what the technology can and cannot reliably do, where it fails and what it needs. Both rungs are about knowledge; Why Executives Need AI Literacy made the case for the second.

The third rung is different in kind. A leading executive does not simply know more. They behave differently: they put their authority behind specific work, they ask for proof before they scale, they change how work is organized and they take ownership of the decisions AI now touches. You can be highly aware and personally fluent and still be on the first two rungs, as the Swedish episode shows.

Two cautions make the ladder more useful. It is not a timeline. Many executives have sharp strategic judgment and thin technical literacy; others can explain a language model in detail and have never changed a single process with one. And it is not a score. The point is to find the rung where your next effort pays most, which is the subject of the next chapter.

Four behaviors that change

Leadership on AI is not a personality type. It is a short list of behaviors that can be observed, practiced and improved. Four matter most for an executive.

Why these four, and not others? The list is this program’s framework, not a taxonomy taken from a single study. It was chosen because each behavior matches a practice that the evidence links to results. McKinsey’s March 2025 survey tested 25 organizational attributes against the earnings impact companies reported from generative AI. Three of its strongest signals line up with three of the behaviors. A chief executive’s oversight of AI governance was one of the elements most correlated with bottom-line impact, which is sponsorship at the top. Tracking well-defined KPIs for generative AI solutions had the most impact of the 12 adoption practices tested, which is asking for evidence. And the redesign of workflows had the biggest effect of all 25, which is redesign³. The fourth behavior rests on experiments rather than a survey: people who are held accountable for a decision check automated advice more carefully⁴.

Two cautions apply. The survey findings are correlations from self-reported data, so they show where results and practices go together, not proof that one causes the other. And the list leaves things out on purpose. Using AI yourself, appointing a chief AI officer or choosing a platform can all help, but they are either awareness or means of carrying out the four, not separate behaviors. These are personal behaviors, the things you do in your own calendar, meetings and budget reviews.

Why Executives Need AI Literacy gave the questions to put to a proposal. The behaviors show up in a different set: the questions a leader puts to their own agenda, and the ones each behavior replaces.

The right-hand column is not more technical than the left. It is more specific, and it puts the executive inside the answer. That is the difference between talking about AI and leading with it.

Sponsoring: lending authority, not a name

Every organization has AI enthusiasts. What most lack is a senior person who will choose a few problems, fund them properly, protect the people working on them and remove the obstacles that only a senior person can remove. That is sponsorship, and it is the behavior people below you notice first.

The effect is measurable. In Boston Consulting Group’s 2025 survey of more than 10,600 leaders, managers and frontline employees, the share of employees who felt positive about generative AI rose from 15 percent to 55 percent when they saw strong support from their leaders⁵.

Sentiment is not value, and the survey shows association rather than proof of cause. But the direction is consistent with everything the program teaches about change: people commit to what their leaders visibly spend time and money on. A sponsor’s calendar is a stronger signal than a sponsor’s speech. McKinsey’s finding on chief executive oversight points the same way from the organization’s side, although only 28 percent of respondents said their chief executive oversaw AI governance³.

Asking for evidence: changing the question

AI produces fluent, confident output, and fluency is persuasive. The executive response is not suspicion of AI but a habit of asking what would prove a claim, before the organization commits to it. As Why Executives Need AI Literacy put it, fluent is not the same as correct, and a good demonstration on one task proves little about the next.

A useful evidence question has a recognizable shape.

Compare two versions of the same request. “Let’s try AI in invoice processing” invites a demonstration. “Can AI cut the time to process a supplier invoice by at least 30 percent, from today’s measured baseline, without raising the error rate, so that we can decide by the end of the quarter whether to extend it to all sites?” invites evidence. The second version is illustrative, but its parts are not: a claim, a baseline, a threshold, a guardrail and a decision date.

Call the stance disciplined optimism: optimistic enough to try, and disciplined enough to ask what would prove it. It is also rarer than it sounds. In the same McKinsey survey, fewer than one in five respondents said their organizations tracked KPIs for their generative AI solutions, the practice most associated with bottom-line impact³.

Redesigning work: past the bolt-on

The most common way to adopt AI is to add it to a task as it already exists: a drafting assistant for the same report, a summarizer for the same meeting. That captures some time. It rarely changes results, because the process around the task stays the same. As AI Is Changing Everything showed with the electric motor, the gains came when factories were rebuilt around the new technology, not when motors replaced the old steam engine on the same shaft.

Survey evidence points the same way. In McKinsey’s March 2025 survey, the redesign of workflows had the biggest effect of 25 attributes tested on whether organizations saw earnings impact from generative AI, yet only about one in five had fundamentally redesigned even some workflows³. Redesign is a leadership behavior because only a leader can authorize it. It crosses team boundaries, changes roles and sometimes removes work that someone is proud of.

Owning decisions: a name on every call

Why Executives Need AI Literacy showed why accountability cannot be delegated. The behavior is making that true, decision by decision. When AI starts to draft, recommend or act, it is tempting to treat its output as the decision. That is where leadership most often slips without anyone noticing. The model recommended it, the vendor certified it, the team approved it, and no single person can say why the organization did what it did.

Ownership is not only a matter of principle. It changes how carefully people check. In experiments on automation bias, the tendency to accept an automated recommendation without checking it, Linda Skitka and colleagues found that making participants accountable for their overall performance or their accuracy reduced those errors⁴. People who know they will have to explain a decision look harder at the advice they are given.

This is what the Swedish debate was really about. Asking a chatbot for a second opinion is not the problem. Leaving it unclear whose judgment governs is.

Story: Alcoa, from aware to accountable

The clearest documented example of a leader turning awareness into leadership is not about AI. AI is too young to have produced a ten-year record. It comes from an aluminum company in Pittsburgh.

Before. When Paul O’Neill became chief executive of Alcoa in 1987, the company already took safety seriously. Its record was better than the average for the American workforce. Everyone at Alcoa knew that safety mattered. That was awareness, and the lost-workday incident rate stood at 1.86 cases per 100 employees.

The shift. In October 1987, O’Neill gave his first speech to investors who expected to hear about inventory and margins. He talked about workplace safety instead, said the existing record was not good enough and set a goal of zero injuries, telling them that anyone who wanted to know how Alcoa was doing should look at its safety figures⁶. Then he changed what he did. Under his policy, the president of a business unit had to report any injury to him within 24 hours, with a plan to make sure it never happened again⁷. And he had a system built that posted every injury within 24 hours, with its root-cause analysis and corrective action, so that every plant could learn from it⁶.

After. By the time O’Neill retired as chairman at the end of 2000, Harvard Business School’s Working Knowledge reported, the lost-workday rate had fallen to 0.2 per 100 employees⁸. A 2019 University of Pittsburgh case study gives 0.5, and puts the rise in market value during his tenure at 3 billion to about 27.5 billion dollars, roughly ninefold⁶. The sources differ on the end figure, and safety was far from the only cause of the financial result. On either figure, the rate fell by more than 70 percent.

Read the case through the four behaviors. O’Neill sponsored safety with his own time and authority and said he was prepared to spend whatever it took⁶. He asked for evidence in a form that could drive action: every incident, with its cause, within a day. The root-cause analyses led to redesigned work on the plant floor. And he made ownership personal: the business unit president, not a safety department, made the call to the chief executive. Nothing in the program asks you to treat AI like workplace safety. It asks you to notice that the same priority, held with awareness, produced a good record, and held with leadership, produced a different company.

The four behaviors at a glance

One map brings together the evidence behind each behavior and the modules where Level 1 builds it in depth. Around them, Modules 02 and 05 supply the literacy and the use cases, Module 08 the economics a sponsor needs, and Module 10 tests the behaviors against a technology that keeps moving. AI Leaders vs AI Followers in Module 01 describes the matching moves for a whole organization.

What this means for leaders

The move from awareness to leadership is not a matter of learning more about AI, though you will. It is a matter of changing a few visible habits. Choose a small number of AI efforts and sponsor them with real money and your own time. Replace “can the AI do it?” with a question that names a baseline, a threshold and a decision. Expect AI to change the process, not just speed up the old one, and authorize that change. And make sure that every decision AI touches has a named owner who can explain it, beginning with your own.

Check yourself

1. An executive who uses AI tools every day is already leading on AI.MythFact
2. Adding an AI assistant to an existing task is usually enough to change business results.MythFact
3. In experiments, making people accountable for their decisions reduced automation bias.MythFact
4. Employees’ attitudes to AI are largely independent of what their leaders do.MythFact
5. Alcoa’s safety turnaround started from a poor safety record.MythFact
6. Sponsoring AI well means approving most of the AI proposals that reach you.MythFact

Reflection: what your calendar says

What comes next

This chapter named the move from knowing about AI to leading with it, and the four behaviors that mark it. The natural next question is personal: where are you starting? The next chapter, Your AI Leadership Starting Point, gives you a short self-assessment, so that you can see which rung and which behavior deserve your effort first.