Two months, five cities, two Georgetown executive MBA cohorts, and one question almost nobody could answer yes to.
There is a moment I got to watch about a dozen times this summer, and it never stopped being uncomfortable.
A room full of founders and CEOs. I ask a simple question: raise your hand if the workflows in your business are defined well enough that AI could actually run them.
The hands do not go up. A couple, maybe. Usually none. Then comes the part I find fascinating: the laughter. Nervous, knowing laughter, in every single city. Everyone in the room instantly understands two things at once. That the question matters enormously, and that they cannot answer it.
Across every session this summer, fewer than 2 percent of hands went up.
Who Was in the Room
These were not AI skeptics or late adopters. I taught AI In Action with David Nilssen of DOXA Talent in Denver, Dallas, Philadelphia, Seattle, and Walnut Creek, and assisted Dr. Brooks Holtom with executive MBA cohorts at Georgetown in Dubai and Washington, D.C. Close to 300 leaders in total.
When we surveyed the July rooms, all but one person was already using AI. Most were juggling two, three, sometimes five platforms at once. These are the enthusiasts. The believers. The ones spending real money.
Yet only three in ten could say their company has a clear AI strategy. Not one of them lacked interest. What they lacked was a way to build an AI leadership program worth the name.
Sit with that combination for a second. Nearly everyone is using it. Almost no one can describe the plan. That gap is the story, and it is not a technology story.
Listen to How They Describe the Problem
We asked everyone to name their biggest pain point with AI. Reading the answers back, three of them stopped me cold:
"Getting AI to do what I am looking for it to do."
"Getting it to perform functions as requested."
"Getting the prompts to do what you envision."
Three different people, three different companies, the same sentence. And notice what kind of sentence it is. Each one blames the machine for failing to follow an instruction. But read them again, slower. Every single one is actually describing an instruction that was never written down clearly. What they envisioned stayed in their head. The AI got a vague sentence and a prayer.
When I pushed on this in the rooms, the real reason AI projects fail surfaced everywhere, in the same two forms. Tribal knowledge: how the work gets done lives in someone's head, usually the person who has been there eleven years and gets interrupted forty times a day. And whatever is not in a head is scattered across systems that were never built to talk to each other.
Two coats, one problem. The knowledge your AI needs is locked up. No model on earth can read a room, and none of them can read your veteran ops manager's mind either.
The Lightbulb Moment
Here is the thing I watch land hardest with our students at the AI Officer Institute, and it stings a little every time.
Most of them came in believing they were building agents. What they were actually doing was logging into someone else's platform, uploading their data, configuring a system they will never own, and arriving right back where they started. A little more optimistic than before. Nothing to show for it.
The reason is brutally simple, and you already know it from the other side of the desk.
Hire someone brilliant on Monday. Hand them a task with no documentation, no process, no access to the systems that hold the answers, and nobody to ask. When Friday's work comes back wrong, you do not call them a bad hire. You call it a bad onboarding. You would be embarrassed to call it anything else.
Give a new employee a task with no information to solve it, and they will fail. We accept that instantly about people. We refuse to accept it about AI.
That is why AI projects fail, and that is what the empty show of hands means. Every one of those companies is onboarding a genius into an organization that cannot explain how anything works. Then they blame the genius.
What Has to Exist First
Getting tribal knowledge unlocked is not mysterious. It is work, and it is exactly the work we have been teaching for over a year. Three disciplines, in order:
- ABC: Always Be Cataloguing. This is information architecture: organizing your data, structured and unstructured, so a system can actually reach it. The spreadsheets and the databases, but also the contracts, the call notes, the style guides, the thousand documents scattered across drives. Named, organized, findable. If your AI cannot find it, it does not exist.
- Workflow Design. Write down how work actually moves. Steps, decisions, exceptions, who owns the output, where a human stays in the loop. Fewer than 2 percent of the leaders I met this summer had done this. That is the whole reason their agents disappoint them.
- Code. Prompts are code. HTML is code. Python is code. You do not need to write it yourself anymore. You need to learn how to create instructions that agents can follow, and the agents will write the code for you.
We covered these in depth in the three skills that unlock the other 50% of leadership. If your AI spend is producing shrugs instead of returns, start there.
The Question Nobody in the Room Could Answer
Late in one session, I asked a different question, and this one produced silence instead of laughter.
Your people have been using AI for two years. Which of them are actually good at it?
Not who is enthusiastic. Not who mentions it in standups. Who could you put in charge of an AI program tomorrow, and who could you trust to build one? Because those are two different people. Leading an AI program means mapping workflows, deciding what gets delegated, and owning outcomes. Building one means architecting the information layer and wiring the system together. They are two different seats in the org structure the AI era demands, and companies keep hiring for one while expecting both, then wondering why everything stalls between the strategy deck and production.
Nobody could answer it. Two years in, with real budgets deployed, not one leader could tell me which of their people are verifiably good at the most important new skill in business.
Find Out
That is the problem we built certification to solve. AI Officer certification tells you whether someone can lead an AI program. AI Engineer certification tells you whether someone can build one.
The companies pulling ahead right now do not have better tools than you. They have the same tools. What they have that you might not is knowledge their AI can actually reach, and certainty about which of their people can do the unlocking.
Find out who you have. Then go unlock the rest.