.png)
September 20, 2026
69% of leaders say they use AI at work, and among people who already use it, only 16% meet Microsoft's bar for running real work through agents. That distance is the knowing doing gap. It is where careers get decided over the next three years, and one 30 minute test will show you where you stand in it.
Thursday morning in Washington, DC, at the 21st Annual National Business Leadership Conference, I asked a room full of leaders to do something in silence.
Pick one workflow you did this week, I told them. Now, in your head, explain step by step how you would rebuild it so an agent runs it and you review the output. I don't mean prompt it. I mean rebuild it. Notice the exact spot where your explanation runs out.
I already had a read on the room. As prework, before anyone walked in, we administered a one question poll: think about your last full workday, and pick the one that describes how AI actually fit into it.
Rounded to the nearest point, the room answered like this. 4% hadn't used AI for work that day. 54% had asked it a few questions or drafted something, then did the rest of the day manually. 14% had one or two recurring tasks set up, with their workflows otherwise unchanged. 18% had at least one regular workflow running through AI from start to finish, with them reviewing. 11% said agents and automations did about 80% of their work that day, while they stayed in the loop for the 20% that needs them.
So 96% were using AI, and 29% had rebuilt even one workflow. More than half were in that second group, asking a few questions and then going back to work the old way. It was a small poll from one room, and the people who answer a poll about AI tend to be the ones most comfortable with it. It still drew the same shape as every national survey I was about to show them.
Afterward, the phrase I kept hearing was "just scratching the surface." One woman stayed to talk. In that minute of silence, she told me, she realized she isn't practicing what she preaches, and neither are the leaders above her. She compared it to the hairdresser who makes everyone else look wonderful and shows up with their own hair undone. Then she confided where she works: one of the big technology companies, one that is building AI. She is surrounded by it all day. Everyone around her is in the know, she said, and very few of them know how to truly use it.
I run that test because of what I hear before I ever walk on stage. Nearly every organization we talk to tells us some version of the same story. Everyone is talking about AI. The company has told people to adopt it. The tools are licensed and sitting on every laptop. So the assumption settles in: we are doing this.
Then you look at how people actually use it. Most are asking AI questions the way they once asked a search engine. A glorified Google. Very few have handed it the grunt work, built an agent around a task they repeat, and moved themselves into the reviewer's chair.
"Del dicho al hecho hay mucho trecho." (From the saying to the doing, there is a long stretch.)
My mom used to say it almost as a song, because in Spanish it rhymes: dicho, hecho, trecho.
Researchers gave that stretch a name in 1999. Jeffrey Pfeffer and Robert Sutton at Stanford studied companies full of smart, well informed people and found that talk had quietly become a substitute for action. Reading the document felt like doing the work. Sitting in the meeting felt like making the decision. They called it the knowing doing gap.
I think it bites harder with AI, and the reason is behavioral. For two, three, four decades, most of us were rewarded on one equation: effort = success. Attending the session is effort. Reading the article is effort. Sitting through the demo is effort. All of it feels like progress because effort always has.
Building is different. When you finally hand a task to AI and it works, it feels effortless, and for people raised on effort that feeling is unsettling. If it was that easy, what was I for? Some people decide it is too good to be true and stop trusting it. Some quietly worry they are making themselves obsolete. So they stay where it's comfortable, in the knowing, where the effort still feels like their own.
The equation that fits this era is impact = success. And impact only shows up on the doing side of the gap.
1. Knowing is high, and it is highest at the top. Gallup surveyed more than 22,000 employed adults in the U.S. and reported in January that 69% of leaders use AI at work at least a few times a year, and 44% use it a few times a week or more. Managers came in at 55%. Individual contributors, 40%. Leaders are the heaviest users in the building. Keep in mind what the question measures. It asks whether you use AI. It doesn't ask what you do with it.
2. Doing is rare, even among the people already using it. Microsoft's 2026 Work Trend Index surveyed 20,000 knowledge workers in 10 countries, every one of them already using AI at work. Only 16% qualified as what Microsoft calls frontier professionals. To count, a person had to say yes to all 3 of these:
(1) I use AI agents for complex, multi step work.
(2) I routinely redesign my workflows around what AI does well.
(3) I take part in structured, repeatable AI practices that scale beyond me.
Remember who was answering: people who already use AI, rating themselves. Inside the group that would raise a hand in any room, 84% had not crossed that line.
3. Almost no organization says it has this handled. Deloitte's 2026 Global Human Capital Trends research surveyed more than 9,000 business and HR leaders in 89 countries. In March it reported that 60% of executives now regularly use AI to support their decisions. When respondents were asked where their organization stands on managing what that means for decisions and accountability, only 5% said it is making great progress. Both answers are self reported, which makes them closer to a confession than a measurement. Use at the top is common. Command of it is rare.
So knowing sits around 69% among U.S. leaders. Doing sits closer to 16%, and that is among people who already use AI. Only 5% of organizations say they are making great progress managing it. I have put versions of those numbers on a chart for every kind of organization, in every industry, and the shape never changes.
Why we believe it: Because everything around us says so. The all hands meetings are about AI. The training calendar is about AI. Your team talks about it, your feed talks about it, and the license is right there in your toolbar. When a subject surrounds you, it starts to feel like a skill you have.
Reframe: Proximity is not practice. You don't absorb a capability by standing next to it.
McKinsey put the question to the people with the widest view. In its Superagency in the workplace research, published in January 2025, it found that almost all companies are investing in AI. Only 1% of leaders called their company mature in deploying it, meaning AI is built into the workflows and driving real business outcomes. Nearly every company is near it. Almost none would say it has changed how the work gets done.
Most of us stand on both sides of this one. We nod along in the AI meeting so nobody sees what we haven't built, and we accept the same nod from everyone else at the table. A whole room can agree it is adopting AI while no one in it has rebuilt a single workflow.
What to do: Change the question you ask yourself and the people around you. Retire "are you using AI?" because the answer is always yes. Ask "what runs differently because of you?" One workflow, rebuilt so that AI does the repeatable part and you review the result, is worth more than a year of attendance.
In 2002, two Yale researchers, Leonid Rozenblit and Frank Keil, asked people to rate how well they understood everyday things like a zipper, a flush toilet, and a helicopter. Then they asked for a step by step explanation of how each one works, and had people rate themselves again. The ratings dropped. They called it the illusion of explanatory depth: we believe we understand things in far more detail than we do, right up until we have to explain them.
One detail in that paper matters for you. The illusion was strongest for how things work. For procedures, the familiar how to knowledge of tasks people already do, it disappeared. When people wrote out how to do something routine, their ratings didn't drop at all. That tells you where to look. You can describe your own workflow without any trouble, because you do it every week. What you will struggle to explain is how an agent would run it: what it reads, what sets it in motion, which calls it can make alone, and where it stops and hands the work back to you. That second explanation is the one that counts.
You don't need a new tool for this. The AI platform your company already approved almost certainly includes a way to build an agent or a reusable workflow now. What you need first is a brief that an agent could follow.
Run this 30-minute test:
On stage I told the room that the spot where the explanation runs out is their real FlipFactor™ score. FlipFactor™ is our proprietary diagnostic at FlipWork, and it measures your AI readiness based on how you actually work. The distance between what we believe we can do with AI and what we can show is exactly what the FlipFactor™ diagnostic was built to measure, because confidence hides gaps that a self rating never will. The 30-minute test shows you one gap in one workflow. The diagnostic shows you the full picture of how you work with AI. You can take it at flipfactor.ai.
Prompt to steal:
I want to rebuild one workflow so that an AI agent runs it and I review the output. Interview me about it one question at a time. Don't accept vague answers. When I say something like "then I check it," ask me what I check for and what I do when it fails. Keep going until you could run this without me. Then write a one page brief with the trigger, the inputs, the steps in order, the decision rules, and the points where the work comes back to me for review. Finish with a list of every place my explanation ran out.
1. Answer the question I asked the room. Go back to the five answers near the top of this issue and think about your own last full workday. Pick the true one, not the one you would say in a meeting. Then write down what it would take to move one answer further along.
2. Say it out loud to one person. Pick a colleague and explain how an agent would run one of your workflows. Out loud, step by step. Then trade places and listen to theirs, and when they get to "and then it just handles it," ask what happens next. You will both find your edge in under ten minutes, and neither of you will be alone there.
3. Trade one hour of knowing for one hour of doing. Look at your calendar for the AI webinar, the panel, or the article you saved. Skip one of them. Spend that hour on the first question mark from your test. You already know enough. Knowing has never been what's holding this list back.
When the work starts to feel too easy, write down the impact.
The first time an agent finishes in four minutes what used to take your Tuesday morning, an old voice will tell you it doesn't count. That voice was trained on effort. Answer it with a record. Keep a running note with three things for each workflow you rebuild: what it produced, what it gave back to you in time, and who was better off because of it.
Your effort hasn't disappeared. It moved to the review, where your judgment decides whether the output is right, and to whatever you chose to do with the morning you got back. The note makes that visible, first to you and eventually to the people who evaluate your work.
The title on your badge is evidence of where you work. It is not evidence of what you can do.
I said that to a room of leaders on Thursday, and it is just as true for those of us without the title. Being around AI doesn't count. Sitting in the session doesn't count. Having the tool on your laptop doesn't count. Only what you have built counts.
The woman who stayed to talk works inside a company that builds AI. It took her one minute of silence to see it.
The next three years will reward the people who did something with what they know. Most of us already know plenty.
Now go back to the saying, the way my mom sang it. Dicho, hecho, trecho. The saying, the doing, and the long stretch between them. You cross it one workflow at a time.
♻️ Share this with someone who has sat through every AI session and is ready to build the first thing.
¡Hasta la próxima, un abrazo fuerte! (Until next week, a big hug!)
