What Teachers Are Still For in an AI World

What my AS Economics students asked an AI — and what they saved for me.

I once taught an AS Economics class. They were ambitious. They wanted all the marks.

They were also using AI. They were using it to explain every essay question they got wrong. They were using it for content.

They did not rely on AI to grade their essays, but came to me instead. The lessons where I went over how to write to the CIE prompt, they showed up for. The lessons where I taught content from the textbook, they silently disengaged.

They asked me things like this:

• How do I write a conclusion with good evaluation?
• Can we conclude with it depends?
• What's the difference between AS and A2 evaluation?
• How do examiners award marks?

These are not content questions. These are examiner questions. They could not get the answer from an AI or a textbook. They came to me because I have seen many marked papers over the years. I know how the marks work for this class and for these prompts.

In short, my students re-sorted my job before the school did. Before the AI policy did. Before I did.

The signal you are watching is wrong

Most teachers I talk to are watching the wrong signal. They are watching what AI can do. They are watching the newest update. Openclaw, Claude Cowork, Gamma, Canva.

Another signal that matters is what your students stop asking you for.

A team of researchers in Jordan found a version of the same pattern with university EFL students. While the students used AI for the language work, they still came back to the teacher for identity lessons, for cultural questions, for the work AI could not carry. It's the kind of work that only a human with context can impart on another human. The paper's title is a aptply: they know AI, but they also know us. The students were drawing a line. They did not tell the teacher where the line was. They just drew it.

Your students may have drawn one too. It will be your job to find it.

The trait did not change. The water level did.

I do not think the kind of teacher who is good has changed. We need flexible, intelligent humans who put our kids' best interest at heart. We always did.

What changed was the water level.

People quote a certain Warren Buffett line about investing all the time — only when the tide goes out do you see who has been swimming naked. AI is the tide going out in the classroom. It has not raised the standard. It has removed the cover.

The teacher who reads the PowerPoint and assigns the textbook questions was always doing the weakest part of the job. They are easily replaced by a custom GPT.

The teacher who knew his or her cohort and calibrated to his or her school's unique context was always doing the actual work. They could tell a student why a conclusion was missing a mark, and exactly what they needed to do to make it up. They could also tell if a student felt off for a day, and went out of their way to care. With AI, the value of this work is finally coming to the foreground.

So go ahead, you can finally teach children, not just content.

What is obsolete

So what is obsolete? It is reading the PowerPoint, generic worked examples, generic essay feedback, lesson plans that go to the same place every year — anything that does not depend on who is in the room.

The research has been saying this for a long time. We just did not listen. Paul Black and Dylan Wiliam published a synthesis in 1998 called Inside the Black Box. They reviewed about 250 studies on formative assessment.

They found something specific. The moment-by-moment judgment a teacher makes about a specific student's work produces effect sizes of 0.4 to 0.7 standard deviations on attainment. That is the kind of feedback that closes the gap between where the student is and where they need to be.

The same review found that generic, graded-only feedback produces near-zero effect — sometimes negative.

The work AI does well today is the work the literature has been telling us was weakest. We were doing too much of it too, and we were getting away with it because some feedback appeared better than none. The tide hid bad practice, but no longer will it continue to stay hidden.

How can we make sure our students get targeted support that utilize AI as an assistant, but not replace the human adult in the room?

What was always the whole job

What was always the whole job?

Three answers from people who have looked at this carefully.

Lee Shulman wrote a paper in 1986 called Those Who Understand: Knowledge Growth in Teaching. He argued that expert teaching is not subject knowledge plus generic teaching skill. It is a third thing — pedagogical content knowledge.

This means knowing how a particular topic lands with particular students, their misconceptions, and their prior knowledge. What effective teaching practice looks like is topic-specific as well as cohort-specific. It also builds with experience.

That is what my students were asking me when they came in with their essay questions. They did not need general subject knowledge. AI gave them that. They needed the specific knowledge of how an examiner reads a conclusion in AS Economics. That is pedagogical content knowledge.

A recent piece in The Hechinger Report by Tanishia Lavette Williams puts it in different words — judgment, care and cultural knowledge. The work no algorithm automates because no algorithm knows who is in front of you.

Here is a small example from my own classroom.

I once taught a monetary policy unit as a jigsaw. I split the learning objectives across groups. Each group learned their objective from the textbook and from AI. They came back and taught each other.

Then we ran a debate on policy evaluation. When does the liquidity trap mean expansionary monetary policy will not work? When do managed exchange rates stop monetary policy from doing what the textbook says it should? How should we prioritize evaluative points when writing monetary policy essays?

The content acquisition happened outside the room. AI handled that part. The evaluative judgment happened in the room with me. That is the move from knowing a concept to knowing when it breaks. AI cannot do it yet. Not because it lacks intelligence. Because it does not know my students and what they typically miss. AI will never fully have our context.

Lean into your role as the curator of learning experiences.

The honest objection

There is a 2025 RCT in Scientific Reports from Harvard that makes a case for AI. Researchers built a custom AI tutor for a physics course. They compared it to an in-class active learning condition. Active learning is already one of the best teaching formats we have. The AI tutor produced higher post-test scores in less time. The students reported being more engaged.

Yet we have to remember that we are teaching children. What might work for a university classroom might not work for K-12 students. Furthermore, even if we defer to an AI to increase test performance, what do we miss out on teaching by making the teacher obsolete? Or even scarier, what are we implicitly teaching?

I have not figured this out

I also once asked AI to plan a 90-minute lesson on company growth — the risks and benefits of growing, and why some firms scale while others stay small. AI gave me case studies. It scaffolded the lesson from AO1 to AO3 — knowledge, application, evaluation. It used real companies as examples.

I thought it would be interesting. It was not. The lesson was too structured, too boring, and not differentiated for this cohort.

The kids were normatively engaged when I told them it was an AI-planned lesson. Their commitment was hanging on a norm — the teacher cared enough to design this for us specifically. When the norm broke, so did the engagement.

I had to reteach a lot of those concepts later.

Why did it fail? John Hattie and Gregory Donoghue published a synthesis in Science of Learning in 2016. They argued learning has phases — surface, deep, transfer. Different strategies work at different phases. The AI plan was strong on surface-phase form, but what my students wanted was deep learning, not an AI helping them cover a breadth of knowledge efficiently. They wanted to wrestle with the knowledge, figure things out themselves, and light their own lightbulbs.

It also failed Shulman's test. AI did not have the pedagogical content knowledge of how my students typically misread company-growth examples. Differentiation is exactly the work pedagogical content knowledge does. AI did not have it. I tried to outsource the part of the job that cannot be outsourced. I screwed up.

More important than the above points, by using AI, I accidentally communicated to them that I didn't care. That realization stung.

Reflecting on that lesson, I should have never used AI to replace the lesson. Instead, I now use AI to enrich it. AI will help me research relevant examples, case studies, and organize my sources that go beyond the textbook. I now use AI to maximize my own impact on learning.

What this means for PD

ISC Research's 2025 report says the typical international school holds onto teachers for one to six years. Continued Professional Development (CPD) is a lever for retention. AI integration shows up on every CPD list nowadays.

Most of those PD plans aim too low. They give teachers a prompt library. They teach AI as a tool the teacher uses.

The higher bar is teaching AI literacy so teachers can teach AI literacy to students. Students need to know how to use AI to enhance learning, not degrade it. The jigsaw lesson worked partly because my students already knew how to use AI to acquire content. If they had not, the lesson would have collapsed. Teachers also need to know the most effective ways to use AI for their own workflows. Efficiency matters because it allows teachers to focus on what matters most: the children. More importantly, teachers need not repeat the same mistakes as me.

What I am still working out

I am still working this out. The piece you just read is one round of thinking, not a conclusion.

What I keep coming back to is the silent move my students made before I caught up. They had already decided what they needed me for. They just were not telling me until I asked.

I listen now for what my students stop asking me. That is the part of my job AI just touched. Not getting questions might not mean you've done a good job as a teacher. It might just mean the students have decided ot trust an AI more than you.

A question for you: how do your students behave differently post-AI? What is that telling you?

A note on method: this issue was produced through the co-creation workflow I'm advocating. The idea, the angle, the practitioner observations, the curated sources, and the final wording are mine. An AI assistant calibrated to my voice (through a guide of phrases I've approved and rejected) did the research legwork on sources I selected and drafted from an outline we agreed on.

References

1. Almashour, M., Aldamen, H. A., & Jarrah, M. "They Know AI, but They Also Know Us": Student Perceptions of EFL Teacher Identity in AI-Enhanced Classrooms in Jordan. Frontiers in Education. 2025.
2. Black, P., & Wiliam, D. Inside the Black Box: Raising Standards Through Classroom Assessment. Phi Delta Kappan 80(2): 139–148. 1998.
3. Hattie, J. A. C., & Donoghue, G. M. Learning strategies: a synthesis and conceptual model. npj Science of Learning 1: 16013. 2016.
4. Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports 15: 17458. 2025.
5. McKenzie, L. The International Schools Market in 2025. ISC Research. 2025.
6. Shulman, L. S. Those Who Understand: Knowledge Growth in Teaching. Educational Researcher 15(2): 4–14. 1986.
7. Williams, T. L. OPINION: The new AI tools are fast but can't replace the judgment, care and cultural knowledge teachers bring to the table. The Hechinger Report. November 4, 2025.