Teachers Are Designing Smarter AI Tutors That Actually Check for Understanding

When schools use AI, the fear is often that it becomes a fancy answer machine. This new study, from researchers in Singapore, flips that worry on its head. They ran a workshop where science teachers co-designed "agentic" AI apps—meaning AI that can take steps and guide students on its own. What they found is encouraging: these teachers didn't want AI to spoon-feed facts. Instead, they designed it to act like a thoughtful science instructor: to ask about a student's initial ideas, spot gaps or wrong beliefs, and show teachers evidence of how a learner is thinking as they work through a problem.

The researchers looked closely at four teacher-made designs. One was a tool that checks students' understanding of chemistry before starting new lessons. Another acted as a dialogue coach for physics concepts like kinetic particle theory. A third helped diagnose experimental design skills, and a fourth offered problem-solving scaffolding where students practice applying science to new situations. Each design connected a specific learning problem to a specific kind of AI interaction, and most produced meaningful evidence for the teacher, like "this student thinks heavier objects always fall faster" or "the class is ready to move on to the next chapter."

But there was a catch. Even though teachers clearly understood how to make AI useful, they were less consistent about control and safety. Only two of the four designs made explicit rules for how the teacher would evaluate or override the AI. Only two named safeguards to prevent the app from giving misleading feedback or doing something harmful. In other words, these tech-savvy educators still left some big decisions to "future developers" rather than spelling them out.

The researchers argue that this is a missed opportunity. They propose a simple five-question protocol for educators and developers to use together: What problem is the AI solving? What does the student actually do? What evidence does the AI return to the teacher? Who keeps final authority—the teacher or the AI? And what safeguards protect students? Having teachers answer these questions before any code is written turns app design into a conversation about good teaching, rather than a technology fad. And it makes sure that AI in the classroom stays a tool for teachers, not a replacement for them.