Every EdTech vendor has an AI feature now, which makes it hard to tell what's actually useful from what's a feature checklist item. Here's a grounded breakdown based on what these tools are actually good and bad at today.
Genuinely useful: first-draft generation
Lesson plan outlines, discussion questions, differentiated versions of a reading passage, a first pass at rubric criteria for a new assignment — AI is reliably good at producing a rough draft of something you then edit. The value isn't that the output is perfect, it's that starting from a draft is faster than starting from nothing, for the same reason editing an essay is faster than writing one from a blank page.
Genuinely useful: grading against something you already defined
Grading a stack of submissions against a rubric you wrote is a well-scoped task: read the text, compare it to specific criteria, produce a score and cite evidence. This is exactly the kind of narrow, well-defined task current AI models handle well, especially when the rubric itself is concrete rather than vague (see our guide to writing better rubrics). It's also why AI grading tools work best as a first pass a teacher reviews, not a fully automated pipeline — the review step is what catches the occasional miss.
Overhyped: fully autonomous grading with zero human review
Any product claiming grades go straight from AI to the gradebook with no teacher checkpoint should raise a flag. Models make mistakes, rubrics have edge cases they don't cover, and a student's specific context sometimes matters in ways a rubric can't capture. The realistic, useful version of AI grading keeps a human approval step in the loop — which is why ClassGrade Pilot is built around review-then-apply, not silent auto-submission.
Overhyped: AI as a replacement for knowing your students
Personalization pitches that promise AI will individually tailor instruction for every student, without a teacher in the loop making judgment calls, consistently overpromise relative to what's actually shipped. AI can help generate options faster; it doesn't know a specific kid the way a teacher who's spent a semester with them does.
Genuinely useful: reducing the search/summarize burden
Summarizing a long article to gut-check whether it's appropriate for a unit, generating a quick comprehension check, restating a concept at a different reading level — these are solid uses because they're fundamentally retrieval-and-rewrite tasks, which is a strength of these models.
The pattern worth remembering
The tools that hold up are the ones doing a specific, bounded task you'd otherwise do by hand, with you reviewing the result. The ones that don't hold up are the ones promising to replace judgment entirely. That distinction is a decent filter for any new EdTech AI feature you're evaluating, including this one.