Effective AI tools for education not only need to work technically but also need to fit in the realities of classrooms. Involving teachers in design brings essential knowledge about students, instruction, and classroom practice into decisions about how AI should support teaching and learning.
This was evident in recent co-design activities across two IES-funded efforts: the GENIUS Center and AutoSPIN project, where science teachers worked alongside researchers and developers to test emerging AI tools, identify potential issues, and help shape how these tools could better support students and teachers.
Seeing AI Feedback Through Students’ Eyes
AutoSPIN, a collaboration between ETS and the University of Georgia, is developing AI-supported formative assessments to better support middle school students’ scientific argumentation. During a recent week-long virtual workshop, teachers explored the latest versions of the activities and integrated their perspectives on the student experience into the improvement of AI support, teacher dashboard, and professional learning resources.
In AutoSPIN activities, students receive AI-generated feedback on their scientific arguments to guide revision. Teachers’ input focused on what this experience might feel like from a student’s perspective. For example, students could become frustrated if the AI feedback system repeatedly asks for revisions without clearly signaling their progress or when their response is sufficient. Teachers suggested clearer indicators to help students see their progress and identify which parts of a scientific argument still need attention.
Teachers also considered how the AI feedback could better respond to different students’ needs. They suggested clearer guidance on distinguishing evidence from reasoning, feedback that builds on students’ previous responses, and different levels of support that fit the instructional context. Their ideas gave the AutoSPIN team a richer picture of how AI feedback can support learning while keeping students actively engaged in scientific reasoning.
Bringing Teacher Judgment into the AI Workflow
Similar ideas emerged through EngageAgent, a teacher-facing agent being developed through the GENIUS Center to generate materials that support learner engagement.
The EngageAgent team worked with two groups of teachers at different stages of development. The first group tested an early version during lesson preparation and classroom instruction. Teachers considered when they would use the generated materials, whether the materials were appropriate for their students, and how much they could trust the content. Their feedback led the team to redesign the material generation process using a five-step prompting structure.
A second group of teachers reviewed materials created through the updated process. They evaluated their quality and usefulness and offered specific suggestions for revision. Their comments showed how teachers draw on their professional knowledge to interpret and improve AI-generated resources. The team is now exploring a Revision Agent (a new Agent within the multi-agent EngageAgent system) that would build teacher review and revision into the workflow.
Translating Teacher Insight into Impact
Across both projects, teachers helped the development teams see the tools through the eyes of the teacher and students who would use them. These experiences reinforced the value of creating meaningful opportunities for teachers to review, revise, and contextualize what AI produces. Co-design gives teachers a voice throughout development and helps teams create tools that support teacher agency, professional judgment, and appropriate trust in AI.
