Past the Threshold

On June 24, Microsoft released the third edition of its annual AI in Education report.¹ The survey was conducted by PSB Insights and reached 3,345 students, educators, and school leaders across the United States, the United Kingdom, Australia, Brazil, Japan, and Saudi Arabia.¹ The headline number is the one that has now become unremarkable. Ninety-two percent of students and education leaders, and eighty-eight percent of educators, said they have used AI for school.² Fifty-eight percent of leaders said their schools are already implementing or scaling it.¹

The harder line in the same report is what sits next to that one. Seventy-seven percent of students and fifty-three percent of educators said they have had no formal training in how to use AI for learning.³ Sixty-six percent of educators and fifty-two percent of students want their institutions to offer training monthly or quarterly.³ Eighty-seven percent of educators and leaders said responsible AI use matters for students' futures.¹

The Threshold Already Crossed

What the report describes, in the careful language of a global vendor, is a building whose front door has already opened. The students walked through it, the teachers walked through it after them, and the school is now trying to write the rules for a foyer everyone has already left. The previous adoption questions, whether students would use these tools, whether parents would tolerate them, whether the curriculum could absorb them, have been answered by the simple arithmetic of the survey. The question now is how the school knows what is happening inside the rooms it has stepped into.

This is the moment most quietly difficult to plan for. A school can write a policy fast. It can announce a training schedule, install a license, post a poster, choose a vendor. What it cannot do quickly is build the habit of actually seeing how a fourteen-year-old uses a chat tool to write a paragraph about Reconstruction, or how a tenth-grader uses one to debug a Python loop, or how a college freshman uses one at one in the morning to keep working when no adult is awake.

What the Training Gap is Really About

It is tempting to read the seventy-seven percent figure as a logistics problem. The students used the tool before the school taught the tool, so the school catches up by teaching it. That solves the training gap on paper. It does not solve the underlying problem, which is that schools cannot teach responsible use of a practice they cannot see. The teacher who has not watched a student work alongside a model has only the final paragraph to grade, and the final paragraph is the same finished thing it has always been, hiding its making.

The integrity worries in the report come from this place. Forty-one percent of students and forty-two percent of educators named academic integrity as a leading concern.¹ Those numbers are not really about cheating. They are about not knowing. A teacher who can see only the finished page cannot say with any confidence which sentences came from the student's own pencil, which arrived after a long pause, which were rewritten three times, which were prompted and accepted in one breath. Without that texture, integrity becomes guesswork, and guesswork eventually becomes either suspicion or surrender.

The Shape of a School That Can See

The most useful response to a ninety-two percent number is not faster adoption or stricter prohibition. It is a different kind of record. What would change, in a school that decided to actually see learning happen, is not the policy document or the vendor list. It is what the school chooses to keep. A draft kept alongside the final. The prompt kept alongside the answer. The revision history kept alongside the grade. The pause logged alongside the percentile. The conversation with the model captured and made available to the adult in the room, not as a forensic exhibit but as the ordinary texture of the year.

This is not an argument against training. It is an argument that training only lands when there is something for the teacher to actually train against. A workshop on responsible AI use lands differently in a school where the teacher has just spent the morning reading three of her students' chat transcripts, watching the moments where each child asked for help and the moments where each child closed the tool and tried again on her own. The conversation she has after that workshop is not theoretical. It is about the children whose work she has already seen.

The Year the Report is Describing

Microsoft framed its report as a call to move from experimentation to implementation.¹ The more interesting line in any vendor report, though, is rarely the roadmap. It is the picture of the year the survey was answered inside. That year, in the spring of 2026, was one where almost every adult and almost every child had used a model, and most had been left to figure out how alone.

A school that wants the next survey to read differently does not need a different number for adoption. That number is going to ninety-five, then ninety-eight, and eventually it will stop being measured. What the school can change is what it knows about the rooms underneath the number.

If ninety-two percent of students are already using AI for school, what is the school keeping a record of while they do?