The original title is about designing AI mentors that give control back to students. Let me rewrite this as a punchy mobile feed headline.

Short summary

The article critiques the design of AI mentors like Google DeepMind's ATL Saathi, arguing the key question is whether students retain authorship and know when to seek teacher help. It proposes a hand-back loop where AI offers bounded hints, students predict and test results, and evidence is captured at each step. The author provides concrete response-card fields and evaluation scenarios to test whether students can correct context and reach a teacher without starting over.

• •AI mentors should prioritize student authorship and human handoff over answer delivery
• •A hand-back loop structure ensures students predict, test, and record evidence before reflecting or escalating
• •Evaluation should use real scenarios, not engagement metrics, to assess whether students retain control

Generated with AI, which can make mistakes.