The Atlantic published a provocation this week: bring back the blue-book exam. Handwritten, timed, invigilated. The argument is not merely anti-cheating. It is that staring down blank pages is how students discover what they are actually capable of as thinkers. Simultaneously, a new arXiv paper on estimating time spent on work tasks is mapping exactly which occupational bundles AI can now complete, using the task-based framework economists use to model labor. The two documents read together are a syllabus for the post-AI credential crisis.
The Homework Problem Is Actually an Epistemology Problem
Tech layoffs are accelerating in August. Fast Company's Michael Grothaus reports that Google, TikTok, Etsy, and Zillow have all cut hundreds of roles in the first week of the month alone. The arXiv task-timing paper is, implicitly, a map of which roles are next. The blue-book exam argument responds to a different but related question: if AI can produce a passable essay in seconds, what does a grade certify? The Atlantic's answer is that the process of writing, the blank page, the pressure, the discovery, is the education, not the output. A 2026 arXiv paper on teaching intro AI when the tools can do the homework proposes a student bill of rights in response to exactly this collapse of the assignment-as-assessment model.
The Anti-AI Writing Movement and the Anti-AI Exam
Hyperallergic's required reading column this week flags the anti-AI writing movement as a growing cultural force. It connects to the blue-book revival in a direct way: both are attempts to reintroduce friction as proof. Not proof that you are old-fashioned, but proof that a human nervous system was present and working. Brewster Kahle's argument in his Culture Slop conversation about public AI and the global brain circles this territory: who controls what counts as knowledge, and who gets to verify it?