1. You've said that "jamming AI into everything" isn't the answer, that it's about balance and using it at the right moment. With schools under pressure to look like they're "doing AI," how do you decide when to use it and when to leave it out entirely?

‍

The pressure to look like you're doing AI is real, and it's leading schools to make decisions in the wrong order. They're picking the tool first, then hunting for a problem it can solve. That's how you end up with chatbots no one uses and licences nobody can justify at budget time.

‍

We start the other way round. What's the actual job to be done? Where's the friction? Sometimes AI is the right answer. Often it isn't. A Year 7 student learning to write a paragraph doesn't need a model that'll write it for them. They need to do the cognitive work themselves. That's not a technology problem, it's a pedagogy decision.

‍

The test we use is whether AI changes what's possible or just changes what's quicker. If it's giving a teacher back an hour of marking time so they can have a real conversation with a struggling student, that's the right moment. If it's letting students skip the productive struggle that's actually the point of the task, that's the wrong moment.

‍

Looking like you're doing AI is a vanity metric. Doing it well is harder, and it usually means using it less than you'd expect. The schools getting this right aren't the loudest about it.

‍

2. Safeguarding is often treated as a compliance checkbox when it comes to AI in schools. You're speaking at COBIS on exactly this topic. What are the safeguarding questions that schools aren't asking and should be?

‍

Most schools are asking "is this tool safe?" That's the wrong question. The right one is "what happens when a student uses this in a way we didn't anticipate?" Because they will.

‍

Here's a question schools aren't asking enough: does your Designated Safeguarding Lead know that AI-generated child sexual abuse material is illegal and must be reported with the same urgency as the real thing? The UK's National Crime Agency found 26% of education practitioners didn't. That's a serious gap. A frontline safeguarding lead might not act on the most serious incident they'll encounter this year.

‍

Then there's the slower harm. Sewell Setzer was 14. Adam Raine was 16. Both died by suicide after intense emotional relationships with AI chatbots. We've trained our staff at Cambridge School of Bucharest on chatbot dependency because it has to sit alongside everything else we already watch for. You won't catch it with a content filter. You catch it because a tutor notices a child preferring an app to a friend.

‍

And nobody's asking the wearables question yet. We had a student turn up in smart glasses last term. We knew what to do because we'd already communicated with parents, briefed students, and updated our acceptable use policy. We were ahead of it. But it was still the first time, so there's that moment when policy meets the real world and you find out whether the work you did months ago actually holds up. Most schools haven't done that work.

‍

The UK Department for Education put it bluntly in their early adopters report: the biggest risk is doing nothing. Compliance gives you a policy. Safeguarding gives you a school where the DSL, the IT lead, and the senior team are having the same conversation, and where they've thought about the incident before it lands on their desk.

‍

3. You've noticed that your students, even younger ones, have largely integrated AI into how they work. What does that tell us about how we've been thinking about the student side of AI adoption, and what does a genuinely useful AI education for students actually look like?

‍

We've been having the wrong conversation. Schools have spent two years debating whether students should use AI, while students have been using it every day. The adoption decision was made without us.

‍

I surveyed 193 of our Year 7 to 9 pupils last term. 90% use AI for schoolwork. 54% use it weekly or more. They're using it most for explaining topics they don't understand, revising for tests, and checking their understanding. That isn't lazy. That's autonomous learning. But it's happening without any of the scaffolding we'd put around any other learning tool.

‍

What that tells us is "should they or shouldn't they" was never the real question. The real question is whether they understand what they're using. I asked a Year 10 class who'd been using ChatGPT for months a simple thing: does it know things, or does it predict things? More than half said it knows things. They were using it daily. They thought it was brilliant.

‍

A genuinely useful AI education isn't a unit on prompt engineering or a list of approved tools. It's teaching students how these systems work, where they fail, what their limits are, and when to trust them. It's metacognition. Knowing when you're doing the thinking and when the machine is doing it for you. But that doesn't happen on its own. Teachers need a way to build it into the subjects they already teach, which is the gap I built AILitKit to close.

‍

4. Teacher adoption of new technology tends to follow a familiar arc: initial training, some enthusiasm, gradual drift. You've deliberately built in ongoing workshops, drop-ins, and peer support at Cambridge School of Bucharest. What makes the difference between AI integration that sticks and AI integration that fades?

‍

The arc you're describing is what happens when training is treated as an event rather than a habit. One INSET day in September, a flurry of activity, then the term gets in the way and people drift back to what they know. The teachers aren't the problem. The model is.

‍

What makes it stick is making it part of the texture of the week. Drop-ins where someone can sit with a colleague for ten minutes and solve an actual problem. Workshops that respond to what's coming up in classrooms now, not what was on the development plan in August. Peer support so people learn from the colleague next door rather than waiting for the expert to fly in.

‍

The other thing that matters is permission. Teachers won't experiment if they think the wrong move ends up in a performance management conversation. We've worked hard at Cambridge School of Bucharest to make AI use a low-stakes thing to try. You bring it to a drop-in, we work through it together, and if it didn't work we figure out why.

‍

Integration sticks when it's coaching rather than training, when it's continuous rather than seasonal, and when the culture says it's fine to be the teacher who's still figuring it out.

‍

5. You hold an AI Governance Professional certification at a time when most schools are still figuring out basic AI policies. What does genuine AI governance look like at the school level, and where are most institutions falling short?

‍

A policy isn't governance. A policy is a document. Governance is the system that makes the document mean something.

‍

Most schools have done the policy bit. They've written something, shared it on the staff drive, and ticked the box. Where they're falling short is everything that comes after. Who reviews it when the law changes? Who decides which tools get approved and which don't? Who's accountable when something goes wrong? If you can't answer those three questions, you don't have governance. You have a PDF.

‍

Genuine governance at the school level looks like a small group of people with named responsibilities. A senior leader who owns the strategic decisions. A Data Protection Officer or equivalent who handles the data side. A Designated Safeguarding Lead connected to both. A review cycle that's actually scheduled rather than aspirational. And a clear route for staff and students to flag concerns without it becoming a crisis.

‍

The EU AI Act, the UK regulatory direction, KCSIE updates, all of it is moving faster than most school review cycles. The schools that'll be in trouble in two years aren't the ones without policies. They're the ones whose policies were written once and never revisited. Governance is what turns a policy into something that holds up when it's tested.

‍

1. You've said that "jamming AI into everything" isn't the answer, that it's about balance and using it at the right moment. With schools under pressure to look like they're "doing AI," how do you decide when to use it and when to leave it out entirely?

‍

The pressure to look like you're doing AI is real, and it's leading schools to make decisions in the wrong order. They're picking the tool first, then hunting for a problem it can solve. That's how you end up with chatbots no one uses and licences nobody can justify at budget time.

‍

We start the other way round. What's the actual job to be done? Where's the friction? Sometimes AI is the right answer. Often it isn't. A Year 7 student learning to write a paragraph doesn't need a model that'll write it for them. They need to do the cognitive work themselves. That's not a technology problem, it's a pedagogy decision.

‍

The test we use is whether AI changes what's possible or just changes what's quicker. If it's giving a teacher back an hour of marking time so they can have a real conversation with a struggling student, that's the right moment. If it's letting students skip the productive struggle that's actually the point of the task, that's the wrong moment.

‍

Looking like you're doing AI is a vanity metric. Doing it well is harder, and it usually means using it less than you'd expect. The schools getting this right aren't the loudest about it.

‍

2. Safeguarding is often treated as a compliance checkbox when it comes to AI in schools. You're speaking at COBIS on exactly this topic. What are the safeguarding questions that schools aren't asking and should be?

‍

Most schools are asking "is this tool safe?" That's the wrong question. The right one is "what happens when a student uses this in a way we didn't anticipate?" Because they will.

‍

Here's a question schools aren't asking enough: does your Designated Safeguarding Lead know that AI-generated child sexual abuse material is illegal and must be reported with the same urgency as the real thing? The UK's National Crime Agency found 26% of education practitioners didn't. That's a serious gap. A frontline safeguarding lead might not act on the most serious incident they'll encounter this year.

‍

Then there's the slower harm. Sewell Setzer was 14. Adam Raine was 16. Both died by suicide after intense emotional relationships with AI chatbots. We've trained our staff at Cambridge School of Bucharest on chatbot dependency because it has to sit alongside everything else we already watch for. You won't catch it with a content filter. You catch it because a tutor notices a child preferring an app to a friend.

‍

And nobody's asking the wearables question yet. We had a student turn up in smart glasses last term. We knew what to do because we'd already communicated with parents, briefed students, and updated our acceptable use policy. We were ahead of it. But it was still the first time, so there's that moment when policy meets the real world and you find out whether the work you did months ago actually holds up. Most schools haven't done that work.

‍

The UK Department for Education put it bluntly in their early adopters report: the biggest risk is doing nothing. Compliance gives you a policy. Safeguarding gives you a school where the DSL, the IT lead, and the senior team are having the same conversation, and where they've thought about the incident before it lands on their desk.

‍

3. You've noticed that your students, even younger ones, have largely integrated AI into how they work. What does that tell us about how we've been thinking about the student side of AI adoption, and what does a genuinely useful AI education for students actually look like?

‍

We've been having the wrong conversation. Schools have spent two years debating whether students should use AI, while students have been using it every day. The adoption decision was made without us.

‍

I surveyed 193 of our Year 7 to 9 pupils last term. 90% use AI for schoolwork. 54% use it weekly or more. They're using it most for explaining topics they don't understand, revising for tests, and checking their understanding. That isn't lazy. That's autonomous learning. But it's happening without any of the scaffolding we'd put around any other learning tool.

‍

What that tells us is "should they or shouldn't they" was never the real question. The real question is whether they understand what they're using. I asked a Year 10 class who'd been using ChatGPT for months a simple thing: does it know things, or does it predict things? More than half said it knows things. They were using it daily. They thought it was brilliant.

‍

A genuinely useful AI education isn't a unit on prompt engineering or a list of approved tools. It's teaching students how these systems work, where they fail, what their limits are, and when to trust them. It's metacognition. Knowing when you're doing the thinking and when the machine is doing it for you. But that doesn't happen on its own. Teachers need a way to build it into the subjects they already teach, which is the gap I built AILitKit to close.

‍

4. Teacher adoption of new technology tends to follow a familiar arc: initial training, some enthusiasm, gradual drift. You've deliberately built in ongoing workshops, drop-ins, and peer support at Cambridge School of Bucharest. What makes the difference between AI integration that sticks and AI integration that fades?

‍

The arc you're describing is what happens when training is treated as an event rather than a habit. One INSET day in September, a flurry of activity, then the term gets in the way and people drift back to what they know. The teachers aren't the problem. The model is.

‍

What makes it stick is making it part of the texture of the week. Drop-ins where someone can sit with a colleague for ten minutes and solve an actual problem. Workshops that respond to what's coming up in classrooms now, not what was on the development plan in August. Peer support so people learn from the colleague next door rather than waiting for the expert to fly in.

‍

The other thing that matters is permission. Teachers won't experiment if they think the wrong move ends up in a performance management conversation. We've worked hard at Cambridge School of Bucharest to make AI use a low-stakes thing to try. You bring it to a drop-in, we work through it together, and if it didn't work we figure out why.

‍

Integration sticks when it's coaching rather than training, when it's continuous rather than seasonal, and when the culture says it's fine to be the teacher who's still figuring it out.

‍

5. You hold an AI Governance Professional certification at a time when most schools are still figuring out basic AI policies. What does genuine AI governance look like at the school level, and where are most institutions falling short?

‍

A policy isn't governance. A policy is a document. Governance is the system that makes the document mean something.

‍

Most schools have done the policy bit. They've written something, shared it on the staff drive, and ticked the box. Where they're falling short is everything that comes after. Who reviews it when the law changes? Who decides which tools get approved and which don't? Who's accountable when something goes wrong? If you can't answer those three questions, you don't have governance. You have a PDF.

‍

Genuine governance at the school level looks like a small group of people with named responsibilities. A senior leader who owns the strategic decisions. A Data Protection Officer or equivalent who handles the data side. A Designated Safeguarding Lead connected to both. A review cycle that's actually scheduled rather than aspirational. And a clear route for staff and students to flag concerns without it becoming a crisis.

‍

The EU AI Act, the UK regulatory direction, KCSIE updates, all of it is moving faster than most school review cycles. The schools that'll be in trouble in two years aren't the ones without policies. They're the ones whose policies were written once and never revisited. Governance is what turns a policy into something that holds up when it's tested.

‍