Ben Williamson asked that question in July 2026. When I checked it against my own practice this week, the data backed him up.
“Bullshit tasks,” in the sense he’s using it, aren’t worthless tasks. They’re busy work dressed up in productivity language, work that keeps you moving without moving you forward. AI is clearly being used for these tasks. The real question is whether schools are about to automate the wrong thing: not teacher labor, but teacher theatre.
The vendors’ pitch is straightforward: AI saves time, reduces workload, and lets teachers focus on the human parts of teaching. Every new education tech company leans into this frame because it’s appealing, and it’s appealing because it’s half-true, which is exactly the problem.
A February 2026 study from UC Berkeley and Harvard Business Review tracked 200 employees across 8 months and found that 83 percent reported AI increased their workload rather than decreasing it, with workdays stretching to 12 hours. The mechanism explains why: AI doesn’t reduce the time a task takes. It lowers the barrier to starting one, so the blank-page fear disappears and people take on more scope, more projects, and more complexity. They get faster at saying yes, not at saying no.
This isn’t new. Ruth Schwartz Cowan documented the same pattern in 1985 with washing machines: labor-saving technology didn’t reduce housework time, it raised cleanliness standards, and everything got washed more often until the time saved upstream disappeared into new expectations downstream. Differentiation is following the same path. It used to be selective because it took real time to build, and now that AI can generate three reading levels or a scaffolded set of notes in minutes, the expectation shifts from “when I have time” to “because the tool makes it possible.” The tool made it easy, so it became mandatory.
A June 2026 District Administration survey shows teachers already understand this: 62 percent use AI, but they segregate it, with 54 percent using it weekly for admin tasks and only 23 percent for instruction. Three in five say AI is eroding trust with students, and two in five now require more handwritten, in-class work as pushback. They’re voting with their feet while districts are still drafting policy.
Stanford’s SCALE report found 800 published studies on AI in K-12 education, and only about 20 offer rigorous causal evidence. That means we’re speculating at scale about long-term effects on deep learning, metacognition, and independence, and the absence of data isn’t slowing adoption down at all.
I tested this against my own work this week while building an EAL-scaffolded packet for AP World History Unit 1. I used AI to evaluate my own materials against 15 EAL teaching strategies, and it found gaps in my coverage, flagging places where my instructions weren’t clear enough for emerging bilinguals. My judgment stayed at the center of the work; AI just extended it.
It would have stopped working the moment I asked AI to generate the full unit packet for me to audit instead. That’s the trap Williamson warns about, where hours of quality-checking and alignment-proofing replace the writing I never did. The time saved upstream just gets buried in checking downstream, so the workload hasn’t gone down; it’s only gotten harder to see.
Claude for Teachers, MagicSchool, and Flintk12 aren’t built for bullshit tasks. They’re built to amplify judgment and extend a teacher’s capacity to think and plan faster. But institutions are wiring them into workflows that assume the outputs are production-ready, when they’re just starting points that still require a thinking human. If you haven’t budgeted for that human labor, you haven’t saved anything.
Are you using AI to sharpen your own thinking, or to replace it? Are you checking whether this tool saved you time, or just made you busier in a way that’s harder to see?
If you’re thinking through questions like this one, where the line sits between AI that extends your judgment and AI that just moves your labor somewhere less visible, my book goes deeper. The AI Doesn’t Know Your Students is available on Amazon and at shouldiuse.ai/book.
David Jacobson is a high school history teacher. He writes about AI, education, and the messy intersection of the two at shouldiuse.ai.
Sources
- Ranganathan & Ye, Harvard Business Review & UC Berkeley Haas (Feb 2026)
- TechTrends / Ruth Schwartz Cowan, More Work for Mother (1985)
- District Administration survey (June 2026)
- Stanford SCALE report, The Evidence Base on AI in K-12
