An assistant superintendent has been forwarded twelve tool requests from teachers in the last two weeks. Three are general-purpose chatbots. Two are reading tutors. One is a feedback generator. The rest are things she has never heard of. Before deciding what to approve, she needs a way to sort them by what each one actually does — in front of students, behind teacher work, or somewhere in between.

The catalog below organizes tools by the job they are doing for a teacher or learner, not by vendor. The organizing question is the same one a thoughtful purchasing committee would ask: what work is this thing meant to do, and what safeguards does the work require?

General assistants can help with drafting, planning, explanation, and analysis. Tutoring systems can support practice and questioning. Teacher-workflow tools can speed up planning and feedback drafting. Voice, music, and video tools can support creative work and accessibility. Each role carries different safeguards.

A new category worth surfacing separately is coding agents. Tools like Claude Code, OpenAI Codex, and GitHub Copilot are no longer just autocomplete in an IDE. They read codebases, propose changes across multiple files, run tests, and iterate on errors with limited supervision. For computer-science classrooms, that changes both what is possible to teach and what students will be tempted to outsource. For teachers in non-CS subjects, these tools rarely belong in the lesson plan — but they belong in the conversation about what professional work will look like for graduates.

The voice, music, and video group has the steepest creative payoff and the trickiest safeguards. ElevenLabs can clone a voice from a few seconds of audio; Udio and Suno generate full songs from a prompt; Higgsfield generates short video clips. In classroom use, these tools shine for accessibility, multimodal review, and student creative projects — and they require firm rules about consent, attribution, copyright, and the difference between using a generated voice and impersonating a real person.

A small set of custom GPTs sits behind the catalog above — for rigor, backward design, outcomes alignment, the science of reading, lesson tutoring, and review songs. They are professional learning support, not magic expertise. These GPTs were built by Matthew A. Zinn for his own teacher coaching work and are shared here for educator use.

A good custom GPT can give teachers a practice partner. It can define a concept, review a lesson plan, suggest a probing-question sequence, surface misalignment, or ask reflection questions. The teacher should still verify sources, adapt to local curriculum, and decide what belongs in front of students.

For school leaders, these examples can model how a district might build constrained assistants around approved frameworks. The constraint is the value: a tool that knows the specific framework and asks teachers to reason through it is safer than a generic chatbot pretending to know the local context.

Tools earn or fail their place by what they actually do for students and teachers. The evidence base on AI in education is rapidly maturing, and it is not one-directional. The strongest signals are around access (reading practice, tutoring availability, adoption scale) and teacher time (drafting, planning, IEP first-passes). The most consistent cautionary signals are around cognitive offloading and the gap between productivity and learning.

The findings below are organized as positive correlations, an adoption snapshot, and cautionary signals. Each card names a specific study or report. None of them are endorsements; all of them are evidence to weigh.

For each case in the catalog above, the same questions apply: What data is collected? Who reviews the output? What learning target is served? What happens when the system is wrong? What does the teacher see? What can the student explain without the tool?

The teacher-stress side of this evidence base is still thinner than the student-outcomes side. Time savings on IEP first-drafts, lesson planning, and routine communication are widely reported by teachers using tools like MagicSchool and the custom GPTs above, but rigorous quantitative measurement of teacher workload and burnout reduction is still emerging. Treat any single vendor's productivity claim as a hypothesis to verify in your own building, not as a settled finding.

Seven resources form a core reading pathway for educators going deeper. Each pairs a single book or framework with the question it answers best. Tools date quickly. Ethical reasoning ages better.

The list should also point back into The Examined Classroom's own AI ethics and thought-experiment pages, because those pages are now the site's strongest differentiated contribution.

Research Resources

Broader reading list for philosophy, AI ethics, and education.

AI Ethics

Policy and ethical frameworks for school decisions.

For Educators

Professional-development resources beyond AI.