Beyond AI-Proof: Lead the Shift to Assignments That Make Thinking Visible

Summary for Educators

Based on Andrew Miller
“Is Your Assignment Vulnerable to AI?”, Edutopia, July 28, 2026

🔵 THE BIG IDEA 

Andrew Miller argues that schools are asking the wrong question about artificial intelligence. No assignment is completely “AI-proof”; tasks exist on a spectrum from highly vulnerable to more AI-resilient. The better question is not “Can AI do this?” but “What thinking must students demonstrate?” Generic summaries, explanations, and reports are especially vulnerable because they often reward completion more than reasoning. Miller recommends redesigning assignments through the A.I.M. framework: Authenticity, Intellectual Demand, and Metacognition. Students should address real audiences and local issues, evaluate evidence and trade-offs, justify decisions, and explain how their thinking evolved. The central tension is between surveillance and design. Schools can devote increasing energy to detecting unauthorized AI use, or they can create learning experiences in which students’ reasoning becomes visible throughout the process. Better design will not eliminate inappropriate AI use, but it can make assignments more meaningful, rigorous, engaging, and resistant to outsourcing.

🔵 KEY TAKEAWAYS FOR EDUCATORS

• Place existing assignments on a vulnerability spectrum: highly vulnerable, somewhat vulnerable, or more AI-resilient.

• Increase authenticity through local issues, real audiences, student experiences, interviews, field research, and primary sources.

• Raise intellectual demand by requiring justification, competing perspectives, trade-offs, multiple solutions, and acknowledged limitations.

• Build metacognition into the process through drafts, reflection, self-assessment, and explanations of how thinking changed.

• Design rubrics that reward reasoning, evidence, decision-making, and revision—not simply polished final products.

• Establish clear expectations for when AI is prohibited, permitted, or required to be disclosed and analyzed.

◻️ WHY IT MATTERS 

Generative AI did not create shallow assignments; it made their limitations more visible. Tasks centered on recall, summary, and generic explanation can often be completed without sustained student thinking—with or without AI. Constant surveillance may also damage trust while failing to keep pace with rapidly changing technology. Assignment redesign offers a more durable response. When students investigate authentic problems, make consequential choices, defend reasoning, document their process, and reflect on their learning, teachers gain stronger evidence of understanding. Students gain experiences that develop judgment, agency, communication, and transferable thinking.

🟢 LEADERSHIP ACTION STEPS

✔ Audit frequently used assignments to identify where students can produce acceptable work without demonstrating meaningful thinking.

✔ Convene teacher teams to sort tasks along the AI-vulnerability spectrum and discuss the evidence behind their judgments.

✔ Redesign selected assignments using Authenticity, Intellectual Demand, and Metacognition as common quality criteria.

✔ Clarify schoolwide expectations for acceptable AI use, disclosure, documentation, and attribution.

✔ Protect collaborative planning time for teachers to examine student work, refine rubrics, and share AI-resilient task designs.

🟡 LEADER REFLECTION

If AI can complete one of our assignments without understanding the subject, what does that reveal about the thinking the assignment actually requires from students?

Original Article

------------------------------

Prepared with the assistance of AI software

OpenAI. (2026). ChatGPT (5.2) [Large language model]. https://chat.openai.com