More than 25 education ministers are at UNESCO in Paris this week to adopt a joint statement on sustaining education as a common good in the age of AI, the centerpiece of Digital Learning Week 2026, which runs from September 8 to 11. In most of the schools and organizations those ministers are responsible for, the question was settled a while ago by people who did not wait to be asked. Someone in the English department writes her feedback comments with AI now. The finance officer drafts her board papers with it. Two heads of year have quietly rebuilt how they write reports. Nobody voted on any of it.
So the leadership question this term is not how to get people started. It is what to do about the fact that they already have.
You do not have an experimentation problem
You have an experimentation-only problem: AI use has spread to almost everyone in your organization and changed almost nothing about how the organization works.
The evidence for that gap is now unusually clear. Microsoft's 2026 AI in Education report, published on June 24, 2026 and based on 3,345 responses across K-12 and higher education in six countries, found that 88% of educators have used AI for school purposes, while only 58% of education leaders said their institution was implementing or scaling it, and 53% of educators had received no formal training at all. The Digital Education Council's Global AI in Higher Education Survey 2026, which collected 45,398 responses across 35 countries, puts the same gap from the student side: only 15% of students say AI is integrated into many of their courses, and 43% have seen it integrated into none.
Read those two findings together and you get the actual state of play. Individual use: near-universal. Organizational practice: barely started. The distance between those two numbers is not a training gap. It is a transfer gap, and it is the thing leaders have to close this year.
Why the enterprise playbook does not fit
The advice that dominates this question was written for a different problem, and following it will send you in the wrong direction.
Search for how to scale AI adoption and you will find the same article from a dozen consultancies: run a pilot, prove the return, build a center of excellence, move from pilot to production. That sequence assumes adoption starts as a controlled deployment that leaders push outward from the middle. In schools, trusts, districts, charities and most public organizations, it did the opposite. Adoption arrived from the edges, one person at a time, on personal accounts, before anyone in the building had decided anything.
Which means the two things the enterprise playbook is best at are the two things you do not need. You do not need to persuade people to try it. You do not need to prove it works, because several people in your organization have already proved it to themselves and will tell you so. A stalled pilot and a resistant staffroom are both real problems, and I have written about why AI pilots so often go nowhere and how to bring the AI sceptics with you. Neither is this problem. This is the problem of an organization full of private competence with no shared practice to show for it, which is why adoption is really a leadership problem rather than a technology one.
The Transfer Test
The Transfer Test is the one question I ask about any use of AI that is already working somewhere in an organization: if the person doing it left tomorrow, would the organization still be doing it next month? If the answer is yes, the practice has transferred and you have adoption. If the answer is no, you have a talented individual, and you should stop calling it adoption.
This is my suggested way of thinking about scale, not a tested method, and it is deliberately harsh. Applied honestly it usually reduces a proud list of twenty AI uses to two or three. That reduction is the point. It tells you where the organization actually is, and it stops leadership teams from reporting individual enthusiasm as institutional progress to a board that cannot tell the difference.
The test is also a diagnosis, because a practice that fails it always fails for one of four reasons.
The four transfers
Turning private use into organizational practice takes four specific transfers, and most organizations only ever schedule the third one.
1. From private to visible
Nobody can adopt what they cannot see, and most good AI work in an organization is invisible on purpose.
People hide it. They hide it because using AI still carries a faint charge of having cheated, because they are not certain it is allowed, and because the first colleague to say "I did that with AI" in a staff meeting is the one who finds out how the room feels about it. Data from outside education makes the same point about time: BCG's fourth annual AI at Work report, published June 2, 2026 from close to 12,000 responses, found that 42% of frontline employees who regularly use AI save around eight hours a week. An organization where a fifth of the staff have each found back a day a week, and none of them has mentioned it, is not a well-run organization. It is one where the safest thing to do with a discovery is keep it.
The transfer is not a showcase event. It is much smaller and much harder: a leader going first, in public, with their own work, including the bit where the output was wrong. What I ask senior teams to do is bring the actual artifact to the meeting. Not a slide about AI. The letter, with the prompt that produced it and the two paragraphs they threw away.
2. From optional to expected
A practice becomes organizational the moment somebody with authority says this is how we do it here, and not one minute before.
Leaders consistently underrate how much work that sentence does. "You might find this useful" is an invitation, and invitations are declined by exactly the people you most want to reach. "From this half term, first drafts of parent letters are written this way, and here is the shared prompt" is a decision. It can be argued with, appealed, and reversed, which is precisely what makes it real.
Note the size of what is being made expected. Not AI in general. One task, named, with a stated standard. The failure I see most often at this transfer is a leadership team trying to make an attitude expected instead of a task.
3. From person to process
The practice has to be written into where the work already lives, not into a course about the practice.
This is the transfer organizations do schedule, and they usually schedule the wrong version of it: a training day. I have trained more than 150,000 educators across more than 30 countries, so I will say this plainly. The training day is not the adoption. The training day creates capable individuals, which you already had. What transfers a practice is putting it inside the template, the agenda, the induction checklist, the handover document, the report proforma. If a new colleague joins in January and picks up the practice without anyone remembering to tell them, it has transferred. If it depends on someone remembering, it has not.
That is also the honest reason so many organizations are stuck. Editing a template is unglamorous. Nobody applauds. And it is the single highest-leverage hour a leadership team will spend on AI this term.
4. From added to replaced
If nothing stopped, nothing was adopted. It was added.
Here is the finding from the BCG report that should worry leaders most: 66% of those frontline AI users receive limited or no guidance on what to do with the time they save, and more than half say they are not reinvesting it in more strategic work. Eight hours a week, found and then quietly absorbed into the day. An organization can run like that for a long time and look busy and improving while getting no benefit it could name to a governor, a trustee or a parent.
So the last transfer is a subtraction. What has come off the list? Which meeting is shorter, which document is no longer written, which two-week task now takes two days and has had the other eight days formally reallocated to something you chose? If a leadership team cannot answer that in specifics, the practice is a hobby with good PR. This is the same discipline behind who should own AI strategy: naming what came off someone's plate is what separates delegation from decoration.
| The transfer | The question to ask | What you see if it has not happened |
|---|---|---|
| Private to visible | Who has seen this being done, other than the person doing it? | Good work discovered by accident, months late |
| Optional to expected | Has anyone with authority said this is how we do it here? | Enthusiastic volunteers and a long tail of non-users |
| Person to process | Is it in the template, the agenda or the induction? | It stops when that person is off sick |
| Added to replaced | What stopped when this started? | Time saved everywhere, time released nowhere |
What I Have Learned From Working With Organizations
The organizations that scale AI are not the ones with the most enthusiasm. They are the ones willing to make a small number of things compulsory.
Across the leadership teams I work with, the pattern is consistent enough that I now look for it in the first hour. The teams that are stuck talk about culture, appetite, readiness and buy-in. They have run awareness sessions. They can name their keen staff. The teams that have moved talk about three or four named tasks, and they can tell you the date on which each one changed, who signed it off, and what was retired to make room.
The uncomfortable part is that the second group usually has less enthusiasm in the building than the first, not more. Making something expected always costs you a little goodwill from the people who liked it better as their own private advantage. That cost is real, and it is worth paying, and I would rather say so than pretend the transfer is painless. Adoption at scale is not the sum of individual enthusiasm. It is a series of small, specific, slightly unpopular decisions that survive the person who made them.
What to do in the next month
Start with what is already working, because you almost certainly have more of it than you think.
- Ask every team to name one thing they now do with AI that they did not do in January. Collect the answers without judgment. You are mapping, not auditing.
- Run each answer through The Transfer Test. If that person left tomorrow, would this still be happening next month?
- For the three strongest, name which of the four transfers has not happened. It will usually be the fourth.
- Do that one transfer, for those three practices only, before half term. Put a date and a name against each.
- Write down what stopped. If nothing stopped, go back to step four.
Nothing on that list requires a new platform, a new post, or a strategy document. It requires a leadership team willing to decide three things in public and retire something. That is the whole of it. The reason it is rare is not that leaders do not know how, but that this work is invisible, unheroic and easy to postpone until a term has gone by.
If your leadership team is holding a real AI-shaped problem and wants to take it from private experiment to something the organization actually does, that is the work I run through Project Momentum, a twelve-week cohort for senior leaders that ends with a decision rather than a report.
Sources and further reading
- AI in Education Report 2026, Microsoft, June 24, 2026. Survey of 3,345 students, educators and education leaders across six countries.
- AI at Work: Why Strategy Matters More Than Tools, Boston Consulting Group, June 2, 2026. Fourth annual edition, close to 12,000 respondents.
- AI in Higher Education Global Survey 2026, Digital Education Council, 2026. 45,398 responses across 35 countries.
- Education in the age of AI: Ministerial statement to be adopted at UNESCO's Digital Learning Week, UNESCO, September 2026.
Dan Fitzpatrick is the founder of The AI Educator and works with schools, trusts, districts and organizations on leading through AI. More about Dan.
