Table of Contents
On a Tuesday night in March, a high school English teacher in Ohio worked through 94 essays. She didn’t read every one the way she used to. She ran them through a screening tool, skimmed the flagged stack, and spent her real attention on the six students whose writing had improved faster than seemed believable. Then she wrote a note to herself about one of them: ask Devon about paragraph three.
That is roughly where things stand. The argument over whether AI belongs in schools is mostly settled, and it did not end with a policy announcement. It ended with a few hundred million people quietly adopting the technology while institutions were still drafting committee reports. ChatGPT launched in November 2022 and reached an estimated 100 million users within two months. Schools found out about it the same way everyone else did.
The Homework Crisis Was Really a Trust Crisis
New York City Public Schools blocked ChatGPT on school networks and devices in January 2023. Los Angeles, Seattle and several other large districts followed within weeks. Within a year most had reversed course, not because the software became safer but because the bans were unenforceable. Students used it on personal phones, on home Wi-Fi, on a friend’s laptop in the back of a bus.
The deeper issue is that homework and grades have always been a proxy for learning, and a fairly loose one. An essay written at 11 p.m. was never strong evidence of much. AI made that weakness visible all at once, to parents and principals and journalists. Schools that responded by redesigning assignments, rather than policing tools, came out ahead. Schools that bought detection software and called it a strategy mostly created suspicion and a pile of false accusations.
Teachers Are Using It More Than Their Principals Realize
Spend an afternoon in a staff room and the honest picture emerges: adoption is running teacher-first, often below the radar. Lesson planning, worksheet variants, quiz drafts, rubrics, parent emails translated into three languages. A 2024 survey from the RAND Corporation found that around a quarter of teachers were already using AI tools regularly, with the heaviest use in English and science. That number has climbed since.
Where the hours actually come back
A middle school science teacher in Bristol described the change to me in plain terms. Planning a week of differentiated lessons used to eat most of Sunday. Now the first draft takes forty minutes, and the rest of the time goes into adjusting the language for the six students with reading support plans. The tool did not replace her judgment. It cleared the admin clutter that used to crowd it out.
Differentiation stops being a fantasy
For decades, every teaching textbook preached differentiated instruction and almost nobody had the hours to do it properly. Producing four reading levels of the same passage, plus a glossary and a set of comprehension questions, was a two-hour job. It is now a fifteen-minute job. That shift matters more than any chatbot demo, because it reaches the students who were previously easiest to leave behind.
Tutoring Now Happens at Midnight
Access is the part of this story that gets undersold. A student in a rural district with one physics teacher and no affordable tutor used to hit a wall at 8 p.m. and stay stuck until Monday. Now the wall has a door in it. Khan Academy’s Khanmigo, school-licensed assistants, and free general chatbots all answer questions at any hour, in any language, without the social cost of admitting confusion in front of a class.
That last point deserves weight. A lot of students who never raise a hand will type a question into a box. The absence of embarrassment is itself a pedagogical feature, and it is not something a human tutor can easily replicate with a fourteen-year-old who is already convinced they are bad at maths.
Assessment Is Being Rebuilt, Awkwardly and Then Better
Take-home essays are in retreat. In their place: in-class writing, handwritten lab notebooks, project portfolios with recorded process logs, and a quiet revival of the oral exam. Several university departments in Australia and the UK have brought back viva-style assessments for exactly the reason you would expect. You cannot outsource a conversation you have to have live.
The return of the spoken answer
Oral assessment was largely abandoned in the twentieth century because it doesn’t scale. Fifteen minutes per student, thirty students per class, and the term is gone. But it is also the hardest format to fake, and a growing number of departments are deciding the trade-off is worth it. Some are using AI to help run the scaffolding around it: generating question banks, flagging likely gaps, producing transcripts for moderation.
The Curriculum Is Shifting Underneath Everyone
Computer science is the clearest case. Introductory programming courses are being restructured around reading and debugging AI-generated code rather than typing it from a blank file, which changes what beginners need to learn first. AI programming is reshaping how coding is taught and who gets to learn it, and the effects are already visible in first-year syllabi.
Beyond that, schools are adding a set of skills that were never on any exam:
- Checking a confident-sounding claim before repeating it
- Writing a prompt specific enough to get a usable answer
- Spotting the moment a polished output is quietly wrong
- Documenting process, not just the finished product
- Deciding when a tool should not be used at all
The credential question follows close behind. If a teenager can learn game design from free tutorials and earn real money before finishing high school, what exactly is a degree certifying? It is telling that more than half of Roblox’s paid creators list high school as their highest level of education. Employers are watching that signal, and so are students.
AI Literacy Is Quietly Becoming a Safety Class
Somewhere between the typing tutor and the coding elective, a new subject appeared: how not to get hurt by this stuff. Schools are now teaching students to recognise synthetic voices, to question a video that seems slightly off, and to understand that a charming stranger online may be a language model with a script. The stakes are not abstract. AI romance scams are rising sharply, and teenagers are prime targets.
The best versions of these classes skip the scare tactics. They hand students a generated image and ask them to find the errors. They run a deepfake of the principal and let the room argue about what gave it away. Detection becomes a skill rather than a lecture.
What Districts Keep Getting Wrong
Three failures show up again and again. The first is buying licences before training anyone, which produces expensive software nobody opens. The second is treating detection tools as evidence, when false-positive rates remain high enough to ruin a student’s term over a writing style that suddenly matured. The third is writing a policy in isolation and handing it down, so teachers comply on paper and improvise in practice.
The schools doing well tend to have one unglamorous thing in common: a teacher or two who was given time to experiment and then time to show colleagues what worked. Adoption spread through the staff room, not through the central office.
The Learning Curve Doesn’t Stop at Graduation
Education is also stretching past the years people traditionally associate with it. Workers retraining mid-career, adults picking up a new trade, professionals who need to stay current in a field that reinvents itself every eighteen months. The institutions paying attention are the ones treating learning as continuous rather than terminal, which is exactly what IEEE’s 2026 Education Week put at the centre of its agenda. The same AI tools that help a fifteen-year-old with algebra are helping a forty-five-year-old switch industries.
Questions Worth Asking Your Local School
If you have a child in the system, or you work in one, the useful questions are specific. Does the school teach students how to use these tools openly, or only punish them for it? Are assignments designed so that a generated answer is obviously insufficient? Do teachers get paid time to learn, or are they expected to figure it out on their own evenings? Is there a clear rule about what student data can be pasted into a chatbot, and does anyone actually know it?
None of those questions have settled answers yet. That is the honest state of play. The technology arrived faster than the research, the research is arriving faster than the policy, and the policy will lag the classroom by years. In the meantime, millions of teachers are making small daily judgment calls about what to keep, what to ban, and what to hand over. Those calls, added up, are what AI in education actually looks like.
