Teaching and Supervision Are Different Jobs

Today's teacher is several jobs in one

A classroom teacher explains content, diagnoses mistakes, prepares exercises, grades work, answers questions, motivates students, manages behavior, communicates with families, and remains responsible for a room full of minors. These functions are packaged together because historically the same adult had to be present to perform the teaching.

That packaging should not be mistaken for a law of nature. If instruction can be delivered reliably by another system, the human functions that remain can be designed separately.

AI has no inherent authority

A student can ignore an AI tutor, close the page, ask it for the answer, or simply stop working. That makes supervision a first-class design problem. Pretending that an AI persona is a strict teacher does not create real authority.

A more realistic model is that families or schools provide the legitimate rules, while software executes them transparently: scheduled study periods, required tasks, mastery checks, progress records, escalation when work is not completed, and clear visibility for the responsible adult.

If instruction becomes autonomous, staffing economics change

Schools currently need many subject teachers because teaching capacity scales with teacher time. If an AI system can independently handle diagnosis, explanation, practice, feedback, assessment, review, and adaptation for each learner, the number of adults needed for instruction no longer has to scale in the same way.

That does not mean schools become adult-free. Supervision, child protection, conflict resolution, physical activities, laboratory work, special needs, social development, and accountability can still require people. But the staffing model could shift from “many adults teaching groups” toward “many AI tutors teaching individuals, with fewer adults supervising and handling exceptions.”

Job reduction would be an outcome, not the measurement target

It is plausible that sufficiently capable AI would reduce demand for some forms of routine instructional labor. It is not responsible to turn a number such as 50% or 90% into a present-day forecast. The useful question is more precise: which teaching tasks can be delegated, for which learners, in which subjects, for how long, and with what level of human oversight?

Those answers can change the labor structure naturally. If one adult can safely supervise a learning environment in which each student receives high-quality individualized instruction from AI, fewer human hours may be required to deliver the same or better instructional output.

Instructional autonomy is the quantity worth measuring

A future evaluation system should move beyond a single tutor score and ask how much of a teaching process an AI can own without a human teacher repairing its work. A possible progression is:

• Assistance: answer questions and provide explanations.
• Guided tutoring: diagnose and adapt within a human-designed lesson.
• Independent lesson delivery: teach a bounded objective end to end.
• Longitudinal adaptation: maintain a learner model and plan across sessions.
• Instructional autonomy: sustain learning outcomes over time with human supervision but without routine human teaching.

The final level is the one that matters for structural change. If it becomes reliable, teaching and supervision no longer need to be performed by the same person.

A research program, not a prediction

The ambition is to make high-quality individualized teaching computationally scalable. Whether that ultimately automates a small share or a very large share of instructional work should be an empirical result, not a slogan.

Teachometry's role is to make the boundary visible: what AI can teach reliably today, what still requires human intervention, and how that boundary moves as systems improve.