The minutes a stopwatch cannot find

Somebody has finally gone and measured it.

An independent randomised trial, published this week, put a stopwatch on primary teachers' lesson planning. Teachers using an AI lesson assistant spent about forty-nine minutes less a week on preparation: two and a half hours instead of three and a third. Independent assessors judged their materials no worse in quality. The evaluators gave the finding a high security rating, which in research terms means they are fairly sure it is real.

Forty-nine minutes. I want to be clear that I am glad of it. Anyone who has watched a teacher eat a sandwich standing over a photocopier knows that forty-nine minutes is not a rounding error, it is a human being getting a piece of their evening back.

But the part of that trial I have not been able to put down is further in.

The tool was built to write lessons from nothing. Given the choice, most teachers did not want that. They used it for about seven per cent of their lessons in an average week, and the use fell away as the weeks went on. Fewer than one in five said it had changed their teaching in any meaningful way. What they kept reaching for instead, the trial found, was help adapting what they already had. The chief executive of the organisation that built it said so himself, publicly: over the term, many chose to adapt existing materials rather than ask for whole lessons from scratch.

So the machine was offered a job, and the teachers quietly handed it a different one.

What is actually in those three hours

Here is what nobody puts in the funding application.

When I plan a lesson, the writing of it is the fast part. I could write you a serviceable lesson on Act One of The Merchant of Venice in twenty minutes, and so could most teachers reading this, and so, now, can a machine. That is not the work.

The work is the second pass. It is sitting with the thing I have written and running it against a specific room. Will the two-page extract defeat the child who reads well but cannot hold the page? Does the group task require four social negotiations before anybody gets to the actual English? Is the instruction discuss with your partner a reasonable request, or is it, for one child in this room, a cliff? Where is the transition, and have I signalled it, or have I assumed that everyone can read my face? Is there a way out of this activity that does not require putting a hand up in front of twenty-eight people?

That pass is not a general skill applied to a lesson. It is a particular knowledge applied to a particular child, and it is the thing that decides whether that child is actually included or merely present.

Inclusion by presence is not inclusion. The difference between those two states is almost entirely made of minutes like these.

Why the stopwatch cannot see them

They are invisible to measurement for an honest reason: they do not look like preparation. They look like a teacher staring at a page.

They are also, inconveniently, not evenly distributed. A lesson needs almost no adaptation for most of a class and a great deal for a few, which means the effort does not scale with the number of pupils, it scales with the distance between them. A measurement system built to count outputs per hour will always read that time as inefficiency. And if we are not careful, a measurement system is exactly what we are about to buy.

Because this is the week it arrived at scale. Free AI training opened to all six hundred and fifty thousand teachers in Ireland and the UK: lesson planning, admin, safer student use, badges at the end. Four and a half million dollars opened for educational technology that can prove it works. Three serious offers inside eight days, and not one of them names accessibility, special educational needs, or a disabled learner. The nearest any of them gets is a passing phrase about diverse learning needs.

I do not think that is malice. I think it is the ordinary shape of a system that was never built for this range of children in the first place. What we run in Ireland is a softened version of an older, narrower, more unbending model: the edges have been sanded, the language has improved, the ethos underneath was never replaced. So when something genuinely useful comes along, it gets fitted to the model we have. The model we have optimises production. So the tool optimises production. And the part of teaching that is not production, the noticing, the anticipating, the quiet redesign of a task so that a child does not have to perform their difficulty in public, gets no tooling, no funding stream, and no line in the trial report.

The thing I would ask for

I am not asking anyone to slow teachers down. God knows.

I am asking that when we buy the forty-nine minutes, we are honest about which forty-nine they are. They are the generic ones. Fine. Take them, with my blessing, and then look at what is left, because what is left is the hard, specific, unglamorous work of making one room fit the actual people in it.

If a tool helped with that, I would be first in the queue. Show me something that reads my own materials back to me and asks the awkward question: this instruction assumes a skill you have not taught; this task has three hidden steps; this is the fourth transition in forty minutes. That is not lesson generation. That is design review, and it is the single most useful thing anybody could put in a teacher's hands.

Change the environment, not the child. We keep saying it, and then we keep buying things that help us produce the same environment faster.

The trial, to its credit, told us the truth without meaning to. Teachers already know where the real work is. They went looking for help with adaptation and found a tool built for generation, so they used it seven per cent of the time and got on with the job themselves: at the kitchen table, after the children were in bed, in the minutes nobody counted.

Elaine Walsh is the founder of Autism & Me (autismandme.ie), a Waterford-based neuroinclusive education consultancy. She writes here on Universal Design, inclusion and Irish education, from the twin vantage points of the classroom and the kitchen table.

Sources: EdTech Innovation Hub, reporting the Education Endowment Foundation trial of Oak National Academy's Aila evaluated by NFER, 7 October 2026. Google for Education blog, "Making AI training available to UK and Ireland educators", 5 October 2026. EdTech Innovation Hub, "2027 Tools Competition opens with $4.5m+ for evidence-based EdTech and AI", 6 October 2026.