We built an AI content generation engine for K-5 learning material (worksheets, practice questions, spelling content). Technically, it could run with zero human review. We don't let it.
Every question, every template, every TTS line goes through teacher vetting before it reaches a kid curriculum itself was co-designed with government school teachers over two years before we wrote any generation code. It's slower and more expensive than pure automation.
We also shipped a simple report button in-app, on the assumption that even a careful human review process will miss things sometimes.
Genuinely asking this community: for content aimed at kids specifically, do you think full AI autonomy will ever be trustworthy enough, or is permanent human-gating just the correct ceiling for this category regardless of how good models get?
[link] [comments]