This guest post, '5 AI Issues Schools Need to Be Aware Of This Year,' is authored by Dan Fitzpatrick, The AI Educator.
There was a point, not that long ago, when a school could reasonably say it was still thinking about AI.
A leadership team could form a working group. Someone could write a policy. A few enthusiastic teachers could experiment with an AI chatbot. Everyone else could watch from a safe distance and wait to see where this was going.
This position will not survive the coming academic year.
Over the past year, I have spent a considerable amount of time with teachers, school leaders, technology leads, and people responsible for AI strategy across schools and districts in different parts of the world.
What interests me most is rarely the polished story. It is what happens once the conference presentation is over. What actually happened when students used the tool? What did teachers stop doing because AI genuinely saved them time? What sounded brilliant in theory but fell apart in a classroom? What are leaders worried about but not necessarily saying publicly? What have schools learned the hard way?
When you listen to enough of those conversations, some fairly clear patterns appear.
These questions sit at the heart of my new book, The Educators' AI Guide 2027, with chapters and contributions from hundreds of educators and school leaders who are already doing this work. The book is valuable because it does not treat AI in education as a theoretical argument. It shows what happens when these questions reach real classrooms, real leadership teams, and real students.
These five trends draw on that wider body of experience. The question for the coming school year is no longer whether schools should engage with AI. Students already are. Teachers already are. The technology is already embedded in the software schools use every day.
The more important question now is whether schools are going to approach this deliberately or continue making decisions one problem at a time.
I think five trends are going to dominate this academic year. They also underpin the rich case studies and expertise you will find in the book.
For many schools, AI still means opening a chatbot, typing a prompt, and waiting for a useful response to appear. Write me a lesson plan. Create ten quiz questions. Rewrite this text for a younger reader. Draft an email to a parent.
That was the first wave. It is not where the technology is heading.
AI is moving toward connected workflows where one action leads into another. Instead of producing a single lesson plan, a system might help develop the lesson, create the resources, differentiate them for different students, build the assessment, and then analyze the resulting work.
Dave Hinrichs and Christie Cloud write about this transition in their chapter, AI Has Moved On. Have Educators? As they put it, "The shift is not simply from one tool to the next. It is from isolated tasks to connected work."
That is a very different proposition. It also reflects how teachers actually work.
Teaching is rarely one isolated task. Planning leads to resource creation. Resource creation leads to adaptation. Adaptation leads to teaching. Teaching produces evidence. Evidence needs to be assessed. Assessment influences what happens next.
The interesting question is therefore becoming less about whether a teacher can write an effective prompt and more about whether they can use AI intelligently across that chain of work.
This creates an immediate problem for leaders. Your staff are probably nowhere near the same level. In the same school, you may have one teacher who has barely touched an AI system, another who uses it several times a week, and another who is quietly building their own tools.
That matters when we think about professional development. A session lasting two hours in which everybody is shown twenty useful prompts might have been reasonable a couple of years ago. It looks increasingly inadequate now.
Schools need to understand where their people actually are. Not where leaders assume they are. Not where the most enthusiastic member of the technology team is. Where they really are.
Then help people move one sensible step forward. For one teacher, that might simply mean becoming confident enough to use AI for planning. For another, it might mean integrating AI into a repeated workflow. For someone else, it may mean designing their own system.
There is no prize for getting everybody to the most advanced stage as quickly as possible. There is also no sense pretending that everybody needs the same training.
Hinrichs and Cloud are careful not to turn this into another judgment of who is keeping up. "The point is not to label who is ahead or behind but to give educators a clearer map of where they are now, what is becoming possible, and what one realistic next step might look like."
If your AI strategy was written a year or two ago, read it again. You may discover it is describing a version of AI that barely exists anymore.
This, for me, is the biggest classroom issue.
AI makes it incredibly easy to produce work. That is useful for adults. It becomes more complicated when the person using it is supposed to be learning how to produce that work themselves.
A student can generate an essay without wrestling with the argument. They can solve a problem without understanding the method. They can summarize a chapter they have barely read. They can produce work that looks remarkably like learning without necessarily doing much learning at all.
In his chapter, The ACTIVE AI Framework, Daren White describes this as the point "where the act of learning is replaced by the act of completing." That is the risk in a sentence. A finished product may look stronger while the learning beneath it has become weaker.
That distinction is going to become harder to ignore. The danger is that schools respond by turning the whole conversation into one about cheating.
I think that would be a mistake. Detection tools will not solve this. Blanket bans will not solve it either.
The more useful question is about when AI enters the learning process.
Imagine two students preparing for a debate. The first asks an AI system to develop the arguments, counterarguments, and evidence before they have properly explored the topic themselves. The second spends several lessons developing their own position, reading sources, and testing their reasoning before using AI to challenge their argument.
Both students have used AI. But cognitively they have done completely different tasks. The second student still owns the thinking.
Samantha Armstrong makes this distinction particularly well in her chapter, AI in the Messy Middle. She writes, "If AI enters before students have a stake in the work, it becomes a shortcut. If it enters after, it becomes a tool."
That is the distinction teachers are increasingly going to have to design for.
Sometimes AI should not be used at all. Sometimes it might be useful for brainstorming. Sometimes students should be encouraged to use it extensively. What matters is that everybody understands which situation they are in and why.
Some teachers are already using very simple systems for this. A task can be labeled red, amber, or green. Red means the thinking needs to happen without AI. Amber means AI can play a limited role. Green means using AI is part of the task itself.
There are plenty of variations, but the underlying principle is what matters. We have to stop treating every use of AI as equivalent.
Students also need a much better mental model of what these systems are. They are not digital professors sitting inside a computer waiting to reveal the truth. They generate responses from patterns in data.
They can be extraordinarily useful. They can also sound completely confident while being wrong. Knowing when to question the system may turn out to be one of the most important academic skills we teach.
There is an equity issue here too.
Giving every child access to a system that can do their thinking for them is not automatically democratizing education. A student who already has strong knowledge and critical thinking skills may use AI to extend their learning. A student without that foundation may use exactly the same tool to avoid developing it.
Armstrong offers an important warning here too: "If we hand them a tool that does the cognitive work, we are not closing a gap. We are widening it."
It is the same technology, but a completely different outcome. That deserves far more attention than it currently gets.
Here is a useful exercise for the start of the school year. Take one of your existing assessments. Give it to an AI model that students can reasonably access. See what happens.
If the system can produce a near-perfect response in seconds, you have learned an important lesson about the assessment.
Larisa Black gives schools an admirably direct test in her chapter, Shifting Assessments in an AI-Enhanced World: "If AI earns 100%, then this assessment probably needs to shift."
That does not mean the subject is obsolete. It does not mean students should stop writing essays. It does mean the final product can no longer be the only evidence we care about.
For years, schools have talked about valuing the learning process. AI may finally force us to mean it.
How did the student arrive at the answer? What changed between the first attempt and the final one? What decisions did they make? Which sources did they trust, and what did they reject? Where did they disagree with AI? Can they explain what they produced?
That evidence suddenly becomes much more interesting.
Black argues that this is where portfolios can become far more useful. Rather than simply storing completed pieces of work, "The portfolio becomes a record of thinking, not just a record of answers."
This does not require every school to redesign its assessment system from scratch. Start small. Collect the plan as well as the essay. Ask students to submit a paragraph from an earlier draft and add a short reflection. Ask them to explain one decision they made. Have a conversation lasting two minutes with a student about their work. Ask what they would change if they had another hour.
These tiny additions make it much harder to separate the finished product from the thinking that produced it.
There is a strange irony here too. AI is creating the assessment problem, but it may also help make better assessment manageable.
Teachers have always known that richer assessment produces better information. The difficulty is time. Portfolios, observations, conferencing, and detailed formative feedback are valuable because they demand human time.
Used carefully, AI can reduce some of the administrative work around that evidence. It can organize notes, structure documentation, surface patterns, and prepare information for a teacher to review.
The important boundary is obvious, and Osman Aliefendioğlu captures it neatly in his chapter, From 21 Hours to 90 Minutes: "The AI proposes. The teacher decides."
We should automate paperwork wherever it genuinely helps. We should be extremely cautious about automating professional judgment. Those are not equivalent.
There was a stage in which the school's AI policy could basically say, "Do not enter personal data into ChatGPT."
That stage has passed. The governance questions are multiplying.
What information is being shared with these systems, and what happens to it? Can a teacher require students to use a particular product? Can parents opt out? Who approves a new AI tool? What happens when a teacher builds their own chatbot? Who has access to student conversations? How are unsafe responses reported? What happens if a tool changes its terms after the school has approved it? Who is actually responsible for checking?
These are not really technology department questions. They are school leadership questions.
The strongest approach I have encountered is usually some form of representative AI group. Not simply the person who loves technology. A strong group should include teachers, senior leaders, people responsible for student safety and data protection, and skeptics. Where appropriate, it should include students and parents, too.
You want people in the room who are willing to ask awkward questions.
Chad Sussex and Sandy Groom-Meeks learned this while establishing their district's AI task force. In their chapter, The Winterset AI Task Force, they write, "While it may be tempting to fill a committee with early adopters, team members who play 'devil's advocate' are invaluable."
Schools also need a proper process for approving tools. Educational value matters. Ease of use matters. Cost matters.
But there should be areas where the answer is simply no. If a product cannot explain what happens to student data, that should matter. If staff cannot see what happens inside a chatbot being used with children, that should matter.
Matthew Wemyss is even more direct in The Chatbot Classroom: "If you are putting a chatbot in front of children and you cannot see what is happening, you already have a problem."
If a tool has weak safety controls, the fact that its lesson generator is impressive is largely irrelevant.
Then there is the emerging complexity of AI companions.
This is going to catch some schools by surprise because much of the use will happen outside school. These products are designed to simulate conversation, friendship, and sometimes emotional attachment. Children may interact with them at home long before a teacher even knows the platform exists. That means digital safety conversations with parents need to evolve too.
We spent years talking about social media. AI relationships are likely to become another part of that conversation. This is where schools need to resist the temptation to write a policy and consider the job finished.
AI governance is going to be ongoing work. Products change. Capabilities change. Risks change. School policies need to change with them.
This is why Sussex and Groom-Meeks argue that success does not come from chasing every new product. It comes from "building a clear, structured framework that prioritizes safety and human connection."
These are not separate issues. Protecting learning, rethinking assessment, and moving beyond policy into shared practice are three parts of the same challenge. They are also the focus of a new webinar series I am leading with 9ine, Protect Learning. Rethink Assessment. Build a Shared Approach.
Across three sessions, we will explore Protecting Learning: When AI Helps and When It Gets in the Way, Rethinking Assessment: What Counts as Student Work Now?, and Moving Beyond Policy: Building a Shared Schoolwide Approach. We will look at how schools can protect thinking and preserve student agency, redesign assessment so it provides better evidence of genuine understanding, and create a shared approach that supports teacher judgment, develops student AI literacy, and gives families genuine confidence.
There are two completely contradictory trends happening in education technology at the moment.
The first is exhausting. There are far too many tools. Every week there seems to be another AI platform claiming it will revolutionize lesson planning, assessment, differentiation, administration, or another task teachers have somehow managed to do for decades.
No teacher can meaningfully evaluate all of them, and they should not try.
Darren Begley writes candidly about this in his chapter, Coping with the Sense of Drowning in New Tools. After trying to keep up with a growing list of AI products, he realized, "I had swapped one list of tasks that was eating up my valuable time for another. I was back at square one."
One of the smartest decisions a school can make this year may be deciding what it is not going to use. Choose a small number of good tools. Make sure they meet your standards. Train people properly. Go deeper.
Constantly chasing the latest release creates another form of workload dressed up as innovation.
But another shift is happening at exactly the same time. Teachers can increasingly build their own tools. That is potentially much more significant.
Until recently, if a teacher wanted a piece of software for a very specific classroom problem, the options were usually limited to three: buy whatever exists, ask the technology team, or forget about it.
That is changing. Software creation is becoming conversational.
Sethi De Clercq explores this shift in Democratization of Custom-Built Tools. In his words, "Software development has become conversational, and the coding language of choice is shifting toward English."
A teacher can describe a problem and increasingly create a functioning tool without being a traditional programmer. It might be a feedback system built around one department's rubric, a revision tool designed around a particular qualification, a small application for tracking information the school's existing system handles badly, or a classroom resource that would never be commercially viable because only thirty students need it.
That changes the relationship between schools and technology.
For decades, De Clercq argues, schools had to adapt themselves to whatever software was available. Now "software can adapt to the needs of teachers instead."
Schools do not have to remain passive consumers. Teachers can become creators. But leaders will need to create the conditions for that to happen safely.
You do not want everybody building random tools using sensitive data with no oversight. You also do not want a governance structure so restrictive that every interesting idea dies before anybody can test it. The opportunity sits between those two extremes.
Create boundaries. Give staff safe places to experiment. Let them build. Let them share what works. And importantly, let teachers learn from other teachers.
There is still a huge gap between the number of educators who have tried AI and the number who feel genuinely confident using it as part of teaching.
I do not think another keynote presentation fixes that. People need examples. They need practice. They need colleagues they can ask stupid questions without feeling stupid. They need ten minutes with a colleague who has already solved the problem they are facing.
That kind of professional learning is less glamorous than announcing a huge AI transformation program. It is also far more likely to work.
If you want practical examples and honest conversations from people already doing this work, the Back To School AI Summit 2026 takes place online on 8 and 9 September. It is free and brings together teachers and school leaders from around the world to share what they are learning while AI is still moving.
This matters at every level of a school, including the earliest years. Tom Millinchip, writing in Rolling Out AI Literacy in Early Years, points out that "students cannot develop meaningful AI literacy if the adults around them are too uncertain to begin the conversation."
Much of what happened over the past couple of years was experimentation. That was appropriate, as nobody knew exactly where this was heading.
Schools tried new approaches. Some worked. Some did not. Teachers explored. Leaders wrote cautious policies. Students ignored some of them.
We learned.
The next stage needs to look different.
The schools that navigate the coming year successfully will not necessarily be the schools spending the most money on technology. I doubt they will be the schools with the longest list of AI tools either.
They will be the schools that know why they are using it. They will have clear boundaries around data and student safety. They will help teachers understand when AI strengthens learning and when it weakens it. They will rethink assessment rather than endlessly trying to police old models. They will give staff space to build expertise gradually. And they will accept that some decisions will turn out to be wrong.
That last part matters. Schools are being asked to make decisions about technology that will continue changing after the decision has been made.
There will not be a perfect moment when someone finally hands educators the definitive AI playbook. We are going to have to build it while the technology is moving.
So start with one step. Look honestly at where your school is today. Choose one problem worth solving. Put sensible safeguards around it.
Try something.
Talk to the people affected by it. Change what does not work. Then move again.
That is the spirit in which The Educators' AI Guide 2027 has been assembled. It is not a definitive playbook. It is a collection of experiences from educators who are thinking, testing, questioning, and sharing while the technology is still moving.
Schools do not need perfect certainty about AI. They need the confidence to make thoughtful decisions without waiting for certainty to arrive.
In the guide, you will find roadmaps, guidance, and practical takeaways from some of the educators who have beaten a path ahead for us.