Two years ago, a worksheet asking students to summarize a chapter and answer five comprehension questions was an ordinary homework assignment. Today, in many households, it is a forty-second task for a chatbot. The question dividing staff rooms is no longer whether that's happening — everyone agrees it is — but what a meaningful assignment even looks like now.
The initial response, in most districts, was detection: plagiarism-checking software retooled to flag AI-generated prose. It has not worked especially well. False positive rates remain high enough that several districts, including two of the largest in Texas, quietly stopped using detection tools for grading purposes altogether last year, citing due-process concerns raised by parents.
A smaller, less publicized shift is proving more durable: teachers are moving take-home work away from tasks that reward polished output and toward tasks that require documented process. Instead of "write a five-paragraph essay on the causes of World War I," some history teachers now assign an annotated outline, a recorded two-minute explanation, or a draft with visible revision history — formats that are harder to fake convincingly and, teachers report, easier to grade honestly.
"I stopped asking for the destination and started asking for the map," said Elena Marsh, who teaches eleventh-grade history in Portland. "A student can hand me a perfect essay generated in a minute. They can't hand me an hour of thinking they didn't do."
"I stopped asking for the destination and started asking for the map."
There is a cost to this shift, and it isn't small: process-based assignments take longer to grade, harder to standardize across a department, and are more vulnerable to accusations of subjectivity from parents used to a rubric with clear right answers. Several teachers interviewed for this piece described spending unpaid summer hours redesigning entire units around this principle, with no additional compensation and no guarantee their district's policy on AI use would still be the same by fall.
What almost no one interviewed believes is that the old model of homework is coming back. Even teachers skeptical of AI in the classroom broadly agree that the take-home essay, as a way of proving understanding, is functionally obsolete for anything a language model can produce convincingly — which, as of this year, is nearly everything assigned before graduate school.
The unresolved question is equity: process-based assessment requires more teacher time per student, and under-resourced schools have less of it. Districts that can't hire enough teachers to grade annotated outlines for 150 students a term may end up defaulting back to easily-automated multiple choice — the one format AI hasn't made obsolete, because it never required much thinking to fake in the first place.
