From Scattered to Strategic: Where Education Really Is With AI Leadership

AI in education feels messy right now: a fast-moving tangle of opportunity and risk.

In every school, every trust, every large education provider, there's a patchwork of experience, understanding and adoption, whether or not leadership has done anything formal about it. A teacher adapts resources with a chatbot on a Tuesday afternoon because it works pretty well, and nobody told her not to. The comms team uses it to draft newsletter copy. A middle leader is increasingly relying on it to pull together summaries and reports. Somewhere an ELT provider is trialling an AI conversation partner for learners who need more speaking practice than the timetable allows, and a maverick enthusiast is building their own AI agent using school data. None of it went through a committee, came up in start-of-term training, or made it into a departmental handbook, but it’s what everyone is doing these days, isn't it?

What I find interesting is how unevenly all this develops, often inside the same organisation. A trust can have governance sorted and no view at all on procurement. A university can have careful, thoughtful work on assessment and nothing on who's accountable for the tools already running in admissions. Almost nobody is behind on everything, and almost nobody has it all solved. That unevenness is a more useful starting point than the usual split between institutions "doing AI well" and everyone else.

The institutions I'd point to as doing this well, in every part of the sector, started by asking what problem they needed to solve. They weren't swayed by the jazzy marketing or by worrying about what the competition down the road was up to. The problem might have been workload, or access, or curriculum outcomes, but the question came first and the shopping came afterwards.

Written down, that sounds obvious. In practice it's rare, because novelty has such a powerful pull on all of us. AI also reached consumers long before it reached institutions, so every organisation already has staff who formed their own opinions and habits with these tools before leadership could define the edges.

Picture a large multi-academy trust that runs a quick staff survey, just to get a sense of the scale of things. In the schools I've spoken to, the finding that worries leaders most is usually about data. Staff are pasting pupil names, assessment notes and sometimes pastoral details into free Ai tools, with no idea where that information goes or who can see it. Nobody has approved this, and nobody has explicitly banned it either. These are good people trying to save time on reports and resources, and very few of them realise they're creating a data protection problem.

The same survey would turn up good practice too: teachers generating retrieval questions or pitching a text at different reading ages, then checking alignment and appropriateness for their pupils. Rather than mandating one platform everywhere and blocking the rest before asking any questions, I would suggest that the better move is to map what's happening and prioritise. First, get pupil data out of free tools and set clear rules on what's never shared. Then cut repetitive admin and spread the good practice that's already there. None of it is flashy, but it’s going to help leadership stay connected to what is really going on.

I worry there's an assumption that every school has an effective AI policy in place by now. The evidence says otherwise. In an NEU survey of 10,311 teachers in English state schools in February 2026, 66% said their school had no policy on pupil use of AI. Universities and colleges look better on paper, with 86% of universities and 49% of colleges offering student-facing guidance on AI use. But guidance isn't the same as an institutional policy, and students still say they're unsure what's allowed, particularly in work that isn't formally assessed.

Where policies do exist, they vary hugely in whether they have an impact on a classroom on an ordinary day. One pattern comes up again and again: a careful, well-meant policy gets written at the top and then becomes irrelevant or impractical when it reaches individual schools or departments, so three leaders give three different answers to the same question in the same week. Clarity really is kindness here. Waiting to see what happens and hoping staff will work it out helps nobody, least of all the teacher who avoids a perfectly good use case because she's afraid of getting it wrong.

The institutions getting this right aren't writing longer or more complex documents. They take the principles they already have and test them against real scenarios in staff meetings and CPD sessions. They bring governors and trustees into the conversation often so that those governors can ask informed questions, and bring meaningful challenge instead of ticking a symbolic box. The stronger examples also tend to have a named senior leader with AI written into their job description. One person can't do all the thinking, but ownership at that level changes how seriously everyone else takes it.

Looking at how AI is being used across institutions and most of it falls into one of three buckets. It replaces something a person used to do. It supplements something already happening and makes it a bit faster or better. Or it does something that wasn't possible before.

The third bucket is where the real change is, and it's also where institutions need to be most careful. It can slide quickly from opportunity into an excuse to give some students a thinner, cheaper version of what better-resourced students still get from a person.

Picture a Year 10 history class where pupils draft an extended answer on the causes of the First World War, hand it in, and wait a week for detailed feedback, because marking extended writing properly takes time. By the time it comes back, much has happened and the class has moved on, so the cognitive wrangle of going back to that writing feels burdensome. An AI tool that gives pupils quick, low-stakes feedback on a draft paragraph immediately changes that. It might point out an unsupported claim or a cause listed without being explained. Without marking the work or replacing the teacher's judgement, it gives pupils more chances to try, get it wrong and try again before the piece reaches the teacher. In the schools I've spoken to, the pupils who gain a lot here are often the quieter or anxious ones, who might struggle to ask for help (especially repeatedly - the fear of overdoing the follow-up questions is very real!) in front of the class. If you try something like this, look for improved iteration and drafting, more confidence and better first submissions. It's worth exploring what the equivalent might be for your own learners. 

Assessment is where we see the most philosophical aspects of the issue collide with the practical , and it's worth us taking a breath. Many institutions have treated AI as a detection or plagiarism problem. I'd argue it's more of a design problem. What is a piece of assessed work meant to show, and what counts as good evidence that a student has learned something, now that AI can produce a polished final product in seconds?

The institutions thinking hardest about this are moving past the ‘catch the cheat’ instinct, and are exploring a combination of two core ideas: assessment security and assessment validity. have landed on something like an antiques dealer's instinct. The emphasis might be less exclusively based on the finished piece and more about its provenance, with artefacts  demonstrating the capability of the student. That trail might include drafts and false starts, a slightly chaotic annotated bibliography, a voice note working through a half-formed idea, or a reflection log about what a student tried and gave up on. You want to see the student wrestling with rough ideas. If a final piece turns up fully formed with nothing behind it, that absence tells you something. Plenty of institutions are capturing this trail with tools already in their tech stack, without buying anything new. Some universities go further and watch the work being built more or less in real time, so a block of text pasted in from nowhere stands out. And then there are the companies offering to fix the problem with more AI. AI marking its own homework? Gosh, it would take a lot to convince me that's the best approach. I keep thinking of the old Irish saying that ‘the hole is more honourable than the patch’. Sometimes the flaws are what make a piece of work feel honest and trustworthy, and human, don’t they? Is there a point where the false injection of those flaws start making cynics of us all…?

The reflex in some places has been to retreat to exam-hall conditions: paper, a pencil, and relief that at least nothing can be faked there. That approach only deals with a small part of the problem. It assumes the worry is a chatbot a student opens on purpose, when AI now sits inside search engines, word processors and tools that don't call themselves AI at all. A policy built around banning devices in a hall does nothing about the work happening everywhere else.

I think there is a more honest and more interesting question coming to the surface: which parts of the curriculum and assessment really matter, and which are busywork wearing the emperor's new clothes of rigour? A worksheet that existed mainly to produce something mindless to tick a box or file away was fairly hollow long before AI came along. Might it even deserve to be undermined? That's an uncomfortable thing to say out loud in a staff meeting. The institutions willing to confront this are using this moment to look hard at the epistemic ownership of ideas and what they ask students to do and why, instead of defending the old ways.

Zoom out, and I think the case for thinking about validity and meaningful co-authorship with AI gets stronger. When anyone can produce a polished piece of writing in seconds, polish alone stops telling you much about the person and their thinking behind it. What employers and universities need is someone who can discern whether that writing is any good and explain their reasoning. Assessment that rewards the thinking behind the work prepares students for exactly that. 

In some ways, we've been here before. For a long time now, teachers haven't been the only place children go for knowledge: books, television and then the internet all changed that, and teaching adapted each time. AI adds more sources and makes them instant, which can feel overwhelming for students and staff alike. If anything, I think that makes a teacher's judgement more valuable. Educators are the makers of meaning, the ones who help students find the signal in the noise.

Ruling things out can be cowardly or brave, a shortcut or a smart decision. From the outside, they might all look exactly the same.

The cowardly ‘no’ is the knee-jerk blanket ban. An institution felt unsure about AI in the early days, blocked it to be safe, and hasn't thought about it since. That's understandable, but it isn't a decision so much as a way of avoiding one.

The brave ‘no’ is more nuanced and contextual, and involves the institution defending a position that might be tricky. Take a school that decides teachers won't use AI to draft end-of-year report comments, because parents read those comments as proof that their teachers know and understand their child. That's a hard line to hold in June, when staff are exhausted and the time-saving is obvious.

I love that the institutions ahead of the curve are becoming much tougher customers. Education hasn't historically been good at IT procurement, and schools are more often told what they need than given the means to question it. Now we are seeing more rigour, more evidence, more impact, and more due process: time-limited pilots, comparing options properly, and looking hard at the evidence instead of taking the vendor's word for it. Hope is not a strategy, especially when it's propping up a decision that was really made by a sales pitch.

My guess is that "AI strategy" won't last much longer as a category of its own. Within a few years it will connect more deeply with procurement, curriculum design, assessment, professional development and operations, much as digital strategy is slowly ebbing away from agendas as a standalone item. The phrase may fall out of use, and "AI" might go with it.

The work won't go anywhere, though. The mess will still be there, and so will the hope and the risk. What will set leaders apart is how they work through it: choosing a purpose before a tool, tightening governance, and redesigning assessment around what it's actually meant to prove. Very little of that is about technology: it comes down to judgement, and whether a school is willing to make deliberate decisions, sometimes unfashionable ones, about where the risks lie and where the opportunities are. That's the thing worth noticing.