The Two-Hour Classroom

At the end of September, twenty-five sixth graders at Kiley Prep in Springfield, Massachusetts will spend their mornings learning from software instead of a teacher.¹ The pilot is an adaptation of the Alpha School model, which asks students to work through their core academics on adaptive AI platforms for about two hours a day and to spend the afternoon on hands-on projects.¹ A veteran teacher will remain in the room during the AI portion, one of the ways this version differs from the private Alpha campuses.²

The Springfield Education Association filed a cease and desist letter last month.³ Its president, Riley Hernandez, told reporters that the union first learned about the plan in August, after a tip.³ The American Federation of Teachers has echoed the concern nationally, arguing that even the most advanced software cannot replace the relationship between a teacher and a student.³ Charles Logan, an education researcher at Northwestern, called the model an open experiment that is not supported by the current research on personalized learning.³

What This Pilot Really Tests

It is a small pilot. One school, one classroom, one grade. Yet it has already become a proxy for a much larger argument that will not stay small for long. If a public school can hand two hours of a child's academic day to a private company's software, and if the results look as impressive as Alpha's marketing suggests, other districts will follow. If the pilot goes poorly, the pushback will shape what public schools are willing to try for years.

Both outcomes assume we will know what actually happened inside that room.

That is the part that keeps getting skipped. Debates about AI in schools tend to argue about which tools should be allowed, which grades should be spared, and who is liable when something goes wrong. They rarely ask a simpler question. When a sixth grader spends two hours on a laptop with a tutor made of language models and telemetry, what will the teacher, the parent, and the school board actually see afterward?

Most AI platforms show a dashboard. Time on task, questions attempted, mastery percentages, a green bar or a red one. That is a summary of activity, not a record of learning. It does not show the moment a student stopped and reread a paragraph. It does not show the two attempts that were almost right and the third that finally caught the idea. It does not show the question the child typed and then deleted before asking. Those small moves are where thinking actually lives, and they are exactly what disappears when learning collapses into a score.

Making the Room Legible

This is why the Springfield pilot matters more than its size suggests. Whatever a school decides to do with AI in the morning, the pedagogical question in the afternoon is the same. Can a teacher walk into that room and understand what each student now knows, how they got there, and where their thinking snagged? If the answer is yes, the software is a tool the teacher can direct. If the answer is no, the teacher is being asked to trust a black box with the intellectual life of twenty-five children.

The future of education will not be settled by a vote on whether AI belongs in the classroom. It will be settled quietly, over hundreds of pilots like this one, by whether the process of learning stays legible to the adults responsible for it. A school that adopts AI without insisting on a visible record of the work is outsourcing more than instruction. It is outsourcing its own ability to notice a child who is drifting, a concept that needs reteaching, or a breakthrough worth celebrating.

Koan is built around that insistence. Every draft a student writes, every revision, every long pause before a sentence, every moment they ask Aidan for help and every moment they close the tutor and finish the sentence alone, becomes part of a record the teacher can read. The point is not that a model watches the child. The point is that a teacher can walk into class the next day and see the shape of the thinking that happened.

The pilot at Kiley Prep will begin in a few weeks. Whatever the twenty-five children learn in those mornings, the more important test will be whether the adults in the building can see it. A school that cannot see its students' thinking cannot teach them, no matter how sophisticated the software running quietly in the background.

So the question for every school leader watching Springfield is not whether to allow AI in the classroom. It is a quieter one. If a child in your school spends two hours tomorrow morning with an AI tutor, what will you know about their thinking by lunch?