Homework is dead.
Not dying. Not “facing disruption.” Dead, in the specific sense that the thing we called homework — a student sits alone, produces an artifact, hands it in, gets a number — no longer proves anything about the student. An educator writing about their own course put it plainly: AI managed to do all of their homework assignments. So they stopped pretending otherwise and rebuilt the course around that fact.
I review AI tools for a living, which means I spend most of my week watching people discover that a product works exactly as advertised and then panic about it. Teachers are the latest group to hit that wall, and honestly, their response is more interesting than most of the enterprise “AI strategy” decks I get sent.
What the fix actually looks like
The change described is almost boringly practical. Written reflections used to be part of pretty much every assignment. Now they’re gone, replaced by in-person interactions with a TA — a scheduled 15-minute conversation after every assignment. Oral exams came back. Written exams came back. Most assignments got redone from scratch.
That is the whole trick, and it’s not a trick at all. If the deliverable can be generated, stop grading the deliverable. Grade the human standing in front of you explaining it.
The broader shift shows up across reporting on this: teachers moving toward projects and discussions to confirm students are actually engaged, and treating homework as a learning process rather than a product. Frontiers, covering homework in the AI era, points at assignment designs like error analysis — hand students AI-generated solutions and make them find and fix the mistakes — and self-generated problems, where students build and solve their own question modeled on class material.
Both of those are, functionally, AI-native assignments. They assume the model is in the room. They just refuse to let the model be the one doing the thinking.
The part nobody wants to say out loud
This educator’s own framing is the most honest line in the whole discussion: AI forced them to abandon evidence-based best pedagogy practices. Not improve them. Abandon them.
That deserves to sit in the open, because the optimistic version of this story — AI frees teachers from grading busywork so they can focus on real learning — skips the cost entirely. Written reflection wasn’t in every assignment by accident. It was there because it works. It got cut not because something better replaced it, but because it stopped being verifiable.
Oral exams are a real assessment method with real strengths. They are also slower, harder to standardize, more vulnerable to a student having a bad day, and brutal for anyone with anxiety or a language barrier. Fifteen minutes per student per assignment is a genuine labor cost, and someone is paying it — usually a TA, usually underpaid.
So the correct description is not “AI improved education.” It’s “AI broke an assessment method and teachers are absorbing the damage with their own time.”
Why this matters outside the classroom
I keep seeing the same pattern in every field I cover, and school is just the clearest version of it.
- Any artifact that can be produced by a model is no longer evidence of the skill it used to signal.
- The only reliable verification left is real-time, interactive, and expensive.
- Organizations that relied on cheap asynchronous proof of work are the ones getting hit first.
Take-home essays. Coding screens. Cover letters. Junior analyst write-ups. Every one of those was a proxy. Every proxy is now negotiable. And the replacement in every case is the same as what these teachers landed on: talk to the person, in real time, about the work.
Hiring already figured this out, badly, by adding more interview rounds. Education is figuring it out more thoughtfully, because teachers actually care whether learning happened, not just whether a box got checked.
My read as a reviewer
I’m not going to tell you this is a happy story. The tools I test are genuinely good at homework, and no detector or honor code is going to change that. What I will say is that the response here is the right shape — not a ban, not a detection arms race, not a policy PDF nobody reads. A structural change that makes the tool irrelevant to the assessment.
The homework-as-process framing is the part worth keeping. If the point is learning, then the artifact was always a receipt, not the goal. AI just made the receipts worthless and forced everyone to look at what they were actually buying.
The uncomfortable follow-up is whether anyone funds the extra hours this takes. In-person verification at scale isn’t a pedagogy problem. It’s a staffing problem wearing a pedagogy costume, and no model release is going to solve it.
🕒 Published: