A teacher knows that a student has been absent for a week, that a family is in the middle of a difficult situation, that yesterday's lesson did not land, or that a usually quiet student finally took a real intellectual risk. None of that may be visible in the assignment sitting on the screen. It is still part of what the work means.
AI systems can process information quickly. They can find patterns, compare a response with a rubric, suggest feedback, and organize a large amount of classroom data. Those abilities may be genuinely useful. But the system sees only what it has been given, while the teacher knows the student, the classroom, the history, and the larger human situation.
That is why I think we need to be very clear about which decisions technology may support and which decisions should remain with teachers. The issue is not whether AI is smart. The issue is whether the decision requires context, responsibility, and human judgment that the system does not possess.
Teachers decide what the assignment is for
Before any AI system becomes involved, somebody has to decide what students are supposed to learn. Is the goal to practice recalling important knowledge, construct an argument, interpret a source, revise for clarity, or demonstrate what the student can do independently? The answer changes what kind of support makes sense.
That is an instructional decision, not a software setting that should arrive preselected. A tool can offer options, explain their consequences, and make them easier to manage. The teacher should decide the conditions under which students will work and make those conditions visible before the assignment begins.
Teachers decide what deserves attention
An AI system may notice that a student opened several hints, changed an answer three times, pasted text into a draft, or repeatedly struggled with a particular kind of question. That information can help a teacher decide where to look. It should not decide what the observation means.
The teacher might respond with a quick conversation, a new example, a small group lesson, another attempt, or no intervention at all. The right response depends on the student and the situation. Turning an observation into a score, warning, or automatic conclusion can erase the very context that makes the information useful.
Teachers decide how to respond
Feedback, assessment, grading, and discipline all carry consequences for students. AI can help organize evidence or suggest a possible next question, but it should not become the hidden authority behind those decisions. Students deserve to know when AI has been involved, what information it used, and where the teacher's judgment begins.
This does not require teachers to distrust everything the system produces. It requires us to treat AI output as advice rather than a verdict. Sometimes the advice will be useful. Sometimes it will miss the point, misunderstand the assignment, or fail to see something the teacher recognizes immediately.
Make room for the relationship
The most important teacher decisions are rarely just acts of classification. They are decisions about what to ask, when to push, when to slow down, what to reteach, what to forgive, and what a particular student may be ready to do next. These decisions are part of a relationship, and relationships require attention and responsibility.
This is the role I want Coach to play. It should help teachers see useful parts of the learning process that are easy to miss when time is short. It should make patterns and possibilities easier to notice. Then it should return the decision to the person who knows the student and is responsible for what happens next.
The promise of educational AI should not be that teachers can finally step out of the way. The promise should be that teachers have better information, more useful options, and perhaps a little more time for the deeply human work that only they can do.