A middle-school science teacher has twenty minutes between an IEP meeting and 4th period to plan a unit on heat transfer. She does not need a Stanford workshop on prompt engineering. She needs a laptop that works, a tool the district has approved, and enough familiarity to ask the assistant for a probing-question matrix and check the result against her standards before the bell. Most defensible classroom AI use looks like that — short, practical, and bounded by the time the teacher actually has.

Schools do not need a custom AI lab to begin responsible teacher-facing use. Basic hardware, stable internet, staff accounts, and digital literacy are enough for planning, brainstorming, drafting, and non-sensitive workflow support. Readiness, at this stage, should be boring on purpose.

The more advanced the use case becomes, the more governance matters. A no-code teacher tool for lesson ideas is one thing. A system that processes student records, recommends interventions, predicts risk, or drafts special-education documents is another. The technology stack is less important than the review stack: privacy approval, clear purpose, teacher training, and a defined human decision-maker.

One of the most immediately useful applications is asking an AI assistant to build a probing-question matrix aligned to a standard, Depth of Knowledge, and Bloom-style categories. The value is not the matrix itself. It is using AI to widen the teacher's planning options before choosing what belongs in the lesson.

Teachers can ask for multiple levels of questions, then review whether the questions actually match the standard, the text, the grade level, and the desired thinking. AI generates the draft matrix quickly. The teacher decides whether a question is recall, application, strategic thinking, or extended thinking.

Khanmigo makes personalization concrete: a student who cares about soccer can meet a polynomial or federalism example through that interest. The deeper point is that AI can help teachers answer the student's old question — why should I care about this?

What the field has learned since the first wave of pilots is that personalization does not automatically produce learning. Many students need help learning how to ask questions, evaluate responses, and stay productively engaged. AI can remember preferences and adjust explanations, but it cannot decide which struggle a student still needs to experience.

The most defensible model is teacher-orchestrated personalization. AI can suggest examples, adjust reading level, vary practice, and offer alternate explanations. The educator decides which path supports the objective rather than distracting from it.

AI can help draft comments, compare work to rubric language, identify unclear feedback, summarize trends, or suggest what a student might try next. That is feedback support. It becomes a different category when the system makes the grade, determines mastery, or becomes the evidence of what a student knows.

Current policy guidance draws this line clearly: AI-generated data can be advisory, but the educator of record determines what the student knows. That boundary is central on The Examined Classroom because it protects both learning and due process.

Administrative uses can also be strong: scheduling, formatting, summarizing non-sensitive information, drafting routine communications, or synthesizing operational data. Leaders still need data privacy review, especially when student records, grades, attendance, disability status, or family information are involved.

AI can save time on IEP work by drafting accommodation ideas, leveling text, generating scaffolded questions, or turning teacher notes into a clearer first pass. For overloaded teachers, that is meaningful support.

But legally and educationally sensitive outputs require a harder line than ordinary lesson planning. AI should not determine eligibility, placement, goals, services, accommodations, or grading decisions for students with disabilities. It can propose language or options for a qualified team to review. The human team must know the student, the law, the setting, and the consequences.

Both truths need to coexist: AI can reduce paperwork friction and improve access, but it cannot replace the expertise of the IEP team or the professional obligation to individualize supports.

Policy & Ethics

Boundaries, detectors, and school decisions

AI Ethics

Frameworks for ethical action

Educator Scenarios

Pressure-test the policy