You can't delegate AI fluency
The answers to four questions tell a board more about its AI transformation than most adoption dashboards:
- Are our most senior people using AI in their own work, or sponsoring it from a safe distance?
- Who in the organization is genuinely fluent, and what is being done to let that fluency spread to others through shared work?
- What has actually changed about how the work gets done?
- Which blockers — such as permissions, incentives, or the fear of getting it wrong in public — are being removed so the new way of working has room to take hold?
These answers rarely sit in program reports. They show up in daily working habits, starting with the leader’s own.
Delegated understanding
Over the past several years, organizations have invested considerable financial resources and human capital into deploying artificial intelligence to achieve better business outcomes and gain a competitive advantage. They’ve made progress. Yet many still aren’t satisfied with where they are on their AI journey. There’s a lot of talk about model choice, operating costs, technical compatibility, workforce training and workflow redesign; all of these are important.
But there is also a fundamental leadership error I see repeatedly, and it deserves a name: delegated understanding. Here’s how it goes: a CEO funds the AI program, appoints an owner, tracks adoption numbers and considers the leadership job done. From the inside, that might look like good governance: delegating the budget, execution and strategy. But it also delegates understanding. Especially at this moment, AI fluency is the one part of the transformation a leader cannot outsource.
Executives can’t, of course, become experts in every technology. That is what specialists are for. But AI doesn’t require advanced training to deploy it for executive work, including analysis, judgment, communication, scenario exploration and organizational design.
A CEO does not need to understand model architecture, any more than they need to prepare tax filings. But every board expects its CEO to be financially literate because financial judgment runs through every decision the CEO makes. We should consider AI fluency the same kind of requirement.
What fluency looks like in practice
Fluency with AI is often discussed as if you either have it or you do not. In practice, it develops in recognizable levels ways, and each level shows up in how someone works, making it observable. The model below offers a synthesis, built as an executive diagnostic rather than a validated assessment, and informed by existing work on AI fluency.
Fluency starts with use
Executives should sit at level 3: using AI as a thinking partner. This means they’ve experienced and internalized what AI can do. It also means they’ve used it to pressure-test their own decisions, game out strategy and test for blind spots. Beyond the value of these outputs, leaders also benefit from knowing whether to trust them, and how to verify outcomes.
This part is important because it means these senior leaders are not merely relying on others to explain what AI is and isn’t capable of. Reaching this level of familiarity does not require technical depth, only the sustained use of AI in the leader’s own decisions until one recurring workflow has genuinely changed.
AI fluency shapes whether leaders use the technology well and what they are able to see. A fluent executive knows to ask what an AI metric actually measures, where the system is likely to fail and what work may simply have shifted out of view. Without that, leaders can easily mistake activity for capability — using “conversations handled”, “calls deflected” or automation rates to justify decisions that later have to be reversed. In mission-critical environments, fluency also includes knowing when not to use AI, what information can be shared with it, how outputs are verified and when uncertain outcomes should be escalated to a person for human judgment. Those behaviors belong at every level above 2. They are part of fluency, not a separate compliance topic.
How fluency spreads
Because AI fluency lives in people, it spreads through contact. Training can create AI awareness, but only shared work creates AI fluency.
The people who accelerate that shift act like seed crystals. Place one genuinely fluent person inside a real workflow, and the people around them start to reorganize how they work, the way a crystal forms around a seed.
But a good seed crystal must demonstrate three characteristics. They must be credible in the work itself, not just in the technology. They must show their process in addition to their output. And they must sit close to the daily workflows where change happens. In the model, these are the level-4 and level-5 people — and it takes a fluent leader to recognize and deploy them.
You can delegate the program, the platform and the training. The understanding must be yours.
Fluency by proximity, not programs
A friend of mine leads corporate services at a large organization. He is not a technologist. One afternoon, I showed a friend who leads corporate services at a large organization how to build an AI coding tool to apply to work he already owned. Though he is not a technologist, he built working prototypes within weeks to support his procurement function — gathering market intelligence and tracking supply chain dynamics. His CEO was surprised as much as impressed, since nobody expected the corporate services leader to transform himself into one of the most AI-fluent people in the company.
Microsoft researchers reported that workplace AI adoption is strongly associated with seeing trusted colleagues use AI in real work. A separate Microsoft study of tens of thousands of its own engineers found that initial use of command-line AI agents spread primarily through social networks. Both studies point in the same direction: colleagues’ making new ways of working visible to others illustrates and encourages the development of AI fluency.
But leaders must create the conditions for that to happen: approving tools that can be used on real work, allowing time to learn during regular work hours, enabling incentives that reward improved ways of working and not just output, and providing the safety to learn — and make mistakes while doing so. Fluency spreads organically under these conditions.
Why leaders hesitate
Senior leaders have spent decades being the most competent people in the room. They are trained to be calm, skeptical and stable, and none of that is a problem. But AI makes everyone a beginner again. So to become fluent in AI, senior leaders must experiment with it personally in addition to sponsoring experimentation across the enterprise.
Here's how I approached this challenge. While leading a complex cloud program, I made a habit of asking a naive question in rooms full of specialists: “I know this might sound out there, but I would bet there is someone in your organization who will say that what I am asking for is doable.” One partner took me at my word, brought in an AI engineer, and together we tested a working AI-enabled approach to cloud migration. Asking cost me little, because I was the senior person in the room.
I had another valuable learning experience when a friend spent a month pushing me to try AI coding tools myself, and I kept finding reasons not to. I did not know how to code, and sitting down at the tool meant facing that directly. When I finally tried, the fear turned out to have been the only real obstacle. That project revealed how the barrier works: the fear is sharpest when the fumbling is your own, and there is no one to delegate to.
This is why, when my corporate services friend needed to build an AI coding tool, I sat with him instead of handing him a list of tools. I knew which approach had worked for me.
What to do now
Each of the four board questions has a matching leadership action.
Every senior leader should be expected to use AI in their own work, because they need to understand how it changes the work they are accountable for. Set level 3 as the standard: one recurring workflow, materially changed, that the leader can describe in concrete terms. Review it the way you would review any other leadership expectation.
Generic training will not drive the change. Find the level 4 and 5 people in the organization, place them inside the workflows that matter, and protect their time so they can work with others rather than becoming an informal help desk.
Licenses bought, courses completed and prompts sent measure activity, not whether work has changed. Ask instead: which decisions are faster or better supported, which process steps no longer exist, which work products are better, and who produced them differently. If those answers are not changing quarter by quarter, the organization is not transforming.
Visible leadership helps remove the fear of getting it wrong in public. Leaders should model the use of AI openly, including the imperfect process of learning. When senior leaders experiment in front of their teams and show how their own work is changing, fluency becomes normal and transmissible. Leading by example is critical to developing organizational AI fluency.
What cannot be delegated
Return to the four questions. Each can be answered through direct experience and evidence, and each begins with the leader’s own behavior. Boards should ask them. CEOs should be able to answer without forwarding the questions to the AI program team.
Delegation is what good executives do. It is how organizations scale. But fluency shapes the judgment leaders use to evaluate the AI-enabled changes for which they remain accountable. You can delegate the program, the platform and the training. The understanding must be yours.
AI is changing the economics of the enterprise
Explore how AI is changing enterprise economics, from cost structures and operating margins to investment risks and board oversight.
What frontier AI means for healthcare leaders
Learn how healthcare leaders can respond to frontier AI cyber risks, protect patient care and strengthen resilience through continuous exposure management.
Sovereignty debt is the new technical debt
As AI and cloud dependencies grow, enterprises face rising sovereignty debt. Learn how digital sovereignty and IT modernization can improve resilience and control.