Ask an instructor how long it takes to mark an essay well and you will get a number between fifteen and forty minutes. Multiply by a cohort. The result is a number that does not fit inside a job, and everyone involved has known this for decades. The response has been a quiet set of compromises: skim reading, comment banks, rubrics applied at speed, marking the introduction and the conclusion and sampling the middle.
None of that is a scandal. It is what a reasonable person does when the arithmetic does not work.
What generative models changed
The compromises worked because a document carried information about its author. A skimmed essay still told you something, because producing five hundred coherent words about Hobbes required having read some Hobbes.
That link is what weakened. The document still exists and still takes the same time to mark, but it no longer certifies the same thing. You are now spending thirty minutes producing careful feedback on a text whose relationship to the student is unknown.
The workload did not get worse. The return on the workload did.
Two responses, one of which does not work
The first response is to add verification on top: run detectors, demand drafts, tighten policy. This adds work to a load that was already over capacity, and the added work has a poor evidence yield. We wrote about why detection specifically is a dead end in an earlier post.
The second is to move some of the assessment weight off the document and onto something the student does in real time. This sounds like more work and is usually less, because a short conversation about a submission is both faster to evaluate and far more informative than a full careful read of a text you cannot attribute.
The numbers we see
A five minute conversation produces a transcript, a summary, and a small number of flagged moments. Reviewing that takes about a minute per student once you have a routine, plus real attention on the few that look unusual. For a class of thirty, that is half an hour to know who understands their own work.
Marking those same thirty essays properly is somewhere between eight and twenty hours. The comparison is not close, and the conversation answers a question the essay can no longer answer at all.
None of this removes the essay. Writing is still worth teaching and worth grading. It removes the pretence that a marked essay, on its own, tells you what the student knows.