The final product doesn’t show the thinking anymore. Design the task so it does.

Navigating Education, The Newsletter. Issue 5, Nov 17, 2026: Who is doing the thinking when students use AI?

Navigating Education | Issue 5: Who’s doing the thinking?

Helping educators and leaders navigate what’s next in teaching and learning.


Who’s doing the thinking? Design the task so it’s the student.

With Claude Fable 5.2 out as of Nov 9, Claude Opus 5.5 out since September, and reverse-engineering skills for AI coding agents now shared on GitHub, I think a lot of the simple software we use in schools can be rebuilt as a lookalike. That’s my view, and it has limits. A lookalike screen isn’t the backend, the data or the learning design underneath, copyright and vendor terms still apply, and Fable’s own safeguards block most cybersecurity work. Earlier this year I wrote Build Your Own EdTech: Why the Future is Local, and I still think schools, districts and teachers will be able to develop their own ed tech and learning software. It’s coming. But that doesn’t mean it’s going to be any good, because we need to focus on the science of learning. A lot of learning and professional development has to happen before someone works like an instructional designer. Without that design knowledge, we get application slop. I wrote more about what reverse engineering really asks of teachers in Reverse Engineering Is Now a Teacher Skill.

So the question under every one of these tools is the same: who’s doing the thinking? A new preprint from Ali Shakiba and Thomas Chaffey at the University of Sydney coded 962 prompts from students’ AI journals in a first-year software engineering course. The course taught students to spell out constraints for the AI, and those prompts were still among the least common, 5.3% of prompts overall. It’s a preprint, not peer reviewed, and it’s one higher-ed software engineering course, not K-12, so we treat it as a lens, not proof. Still, it shows the problem. The finished product doesn’t tell us who did the thinking.

Here’s where I land. We design tasks so students do the thinking. We make that thinking visible through ongoing formative assessment, at checkpoints along the way, not just at the end. And we let AI help build the formative checks and guided practice, with examples and non-examples, and a teacher checks every item before students see it. Start with one unit and one checkpoint. Does that make sense?

Source: Shakiba, A., & Chaffey, T. (2026). Who is doing the thinking? Characterising human agency in first-year students’ AI interactions (Version 1) [Preprint]. arXiv. arXiv preprint


Navigating Education, The Newsletter: Podcast. This week's episode. With the Navigating Education podcast logo.

Podcast

Matt on the left against a soft blurred background. Blue panel on the right reads Episode 82, Start AI on the teacher side, Align it with the science of learning, with the Navigating Education podcast logo.

Start AI on the Teacher Side | Navigating Education Ep 82

When we implement AI, start on the teacher side of the house. There is a very thin veil of studies on AI in K-12 settings, so I don’t think we should spend more time on the student side unless the use is explicit, an extension, or maybe for far older students.

If teachers are using AI to develop materials, they should build them to align with the science of learning: limiting cognitive load, dual coding, worked examples, and spacing and interleaving. These things are very doable with AI. Teachers first. That’s what matters.

Watch the episode


Navigating Education, The Newsletter: From Dr. Matt. New writing across my sites. With the matthewrhoads.com logo.

From Dr. Matt Rhoads

I Can Build Courses in Hours Now. That’s Why I’m Worried About Our Students. AI harnesses let experienced educators build courses in hours, while the same tools can quietly take the learning away from students who don’t yet have the knowledge to check them.

Build Courses in an AI Harness. If you design courses, how much of your time do you actually spend inside an AI harness? The speed is not the point. The point is what the speed is built on.

On Co-Teaching Evolved. How co-teachers divide the work without splitting the partnership: Division of Labor: Making the Invisible Work Visible and Split the Planning: Keep One Shared Plan.

For coaches, on 25 Tips. Plan With AI: Keep the Thinking in the Room.

Books

These are the published books, newest first. Full details are on my books page.

Work with me

Better instruction first. Then we help teachers use AI where it helps. Most partners start with instruction, co-teaching, helping teachers use AI, or coaching their leaders, from a single PD day to a year-long partnership. If that would help your team, start with a 20-minute call: Book a discovery call. The offers are on the consulting page. If your district already has a Solution Tree agreement, you can request me through my Solution Tree profile.


This week across the ventures

Navigating Education, The Newsletter: Evidence Signal. What the research actually says. With the Evidence Signal logo.

Evidence Signal

An AI tutor that lowered grades in same-course sections. A randomized trial with 2,379 undergraduates at a large U.S. public university found final grades 0.37 standard deviations lower in same-course sections given an AI tutor, and only about 15% of students offered it used it even once. In the full sample, which mixed courses, the grade effect wasn’t significant, though estimated losses were larger for first-generation students. Level 3, provisional: a working paper, one university, one semester. Read the weekly roundup

Does AI help make people give up once it’s gone? In randomized trials with 1,222 people, the authors report AI help raised short-term performance, but people did worse and were more likely to give up once it was taken away. Level 3: a preprint, and Evidence Signal read the abstract, not the full paper. Read the explainer

Students skip retrieval practice to protect how studying feels. We already know retrieval practice works. The harder question is why students still avoid it. A new paper by Lan Anh Do and Ayanna Thomas at Tufts, published Oct 26 in the Journal of Experimental Psychology: Applied, looked at that directly.

In three lab experiments, the authors found that people told to protect how studying feels spent less time on retrieval practice than people told to prioritize learning. That held even when time was limited, even after they were told retrieval practice works, and even when they were choosing for a hypothetical learner instead of themselves. In a classroom study with undergraduates, few students chose retrieval practice, and the authors report those choices tracked emotional goals and how effective the strategy felt. They call the classroom results preliminary.

This is peer-reviewed, which is stronger than a preprint, but it’s still adults and undergrads, not K-12, so we shouldn’t treat it as a K-12 classroom finding. The practical takeaway for me is simple. Showing teachers the effect size isn’t enough. We have to make retrieval feel less threatening: low stakes, no grades, just a check on the thinking. Does that make sense?

Source: Liu, J., Sweet, T., Chen, M. H., Engelberg, J., Masters, M. C., Clark, M., Persaud, A., Lancaster, A., Hollingsworth, J. K., & Rice, J. K. (2026). The effects of course-integrated AI tutoring on student performance and engagement: A randomized university trial (EdWorkingPaper No. 26-1598). Annenberg Institute at Brown University. EdWorkingPaper

Source: Liu, G., Christian, B., Dumbalska, T., Bakker, M. A., & Dubey, R. (2026). AI assistance reduces persistence and hurts independent performance (Version 6) [Preprint]. arXiv. arXiv preprint

Source: Do, L. A., & Thomas, A. K. (2026). How learning and emotional goals shape strategic regulation of learning. Journal of Experimental Psychology: Applied. Advance online publication. JEP: Applied

Navigating Education, The Newsletter: AgenticED. AI agent use cases, with the teacher in charge. With the AgenticED logo.

AgenticED

Drafting practice questions, with the teacher in charge. An agent reads the teacher’s unit outline, drafts practice questions with answer keys tied to source passages, and flags any question it can’t ground in the material. The teacher reviews every question, and the use case is rated conditional and on hold, still Under review like every entry in the catalog. Browse use cases

Navigating Education, The Newsletter: EduConference Radar. Open calls for proposals. With the EduConference Radar logo.

EduConference Radar

ETSU Micro-CHIIPs 2027: Teaching for Accessible Learning. A virtual half-day on Jan 6, 2027, looking for session or poster proposals from higher education classrooms. Proposals are due Dec 6, 2026, at 6 PM Eastern. Call for proposals

IETA 2027, Boise, Feb 3 to 5, 2027. The Idaho Education Technology Association call is open to educators, technologists and others with experience to share, members or not, with topics including AI and digital learning. Plan to submit by Dec 31. Call for proposals

Navigating Education, The Newsletter: AI Policy Tracker. K-12 AI policy, state by state. With the AI Policy Tracker logo.

AI Policy Tracker

California’s model policy puts the thinking first. The California Department of Education’s model AI policy for districts and charter schools, posted as required by SB 1288 (2024), is a starting point, not a mandate. It says “human thinking, process, equity, and insight should be emphasized over easily AI-generated products.” California page

A free staff and student AI use checklist. A printable list of questions for a district setting its AI rules, including whether an AI detection result will ever be used on its own to discipline a student or lower a grade. Not legal advice. Get the checklist

Google Search goes off by default for K-12. Google Search, Gemini, Gemini Notebook, and Gemini in Classroom will be off by default for K-12 student accounts until admins turn them on. Google says it will share more details with schools in the coming months before the change rolls out. Districts can keep Gemini off and still turn Search on, but EdSource and K-12 Dive report there is no AI-free Search: AI Overviews and AI Mode stay built in. Common Sense Media, cited by EdSource, found that in 180 tests, AI Mode completed the homework every time. That’s a procurement decision, and it needs a written rationale, not a default click.

Source: California Department of Education. (2026, June 26). Model policy: Artificial intelligence in education. CDE model policy


Close

One move for Monday. Pick one assignment where AI could do most of the work, and add one checkpoint where students show their thinking first.

If someone on your team should be reading this, send it their way: Navigating Education. Want help designing this with your team? Just reply.

— Dr. Matt


Published by Matthew Rhoads, Ed.D.

I'm Dr. Matt Rhoads, an edtech and instructional leader, professor, consultant, and author. I help schools start with how people learn, then bring in AI where it earns its place. Work with me

Leave a Reply

Discover more from Dr. Matt Rhoads

Subscribe now to keep reading and get access to the full archive.

Continue reading