What Orange County Would Not Automate

On July 28, three days before the fall semester was scheduled to start, the Orange County Public Schools board in Florida approved its first policy on artificial intelligence in classrooms.¹ The district serves roughly two hundred thousand students across the Orlando metro. The drafting had taken most of the spring. Each of the seven board members had, along the way, held a town hall with students. Twenty thousand surveys had come back from teachers and parents.¹ The vote passed, though not unanimously.²

The policy takes effect on August 11, the first day back for OCPS students.²

Most of the coverage this week has led with what the policy will not allow. Only district-approved AI tools may run on district networks. No AI detection software may be pointed at student work. No deep fakes. No student data used to train external models.¹² Those are the parts that will make the summaries.

The sentence worth reading twice is inside the section on staff responsibilities.

The Grade The District Kept

The policy says, in a small clause with unusual weight, that employees must remain responsible for decisions involving AI, and that staff members must make high-stakes decisions, including grading, without relying solely on artificial intelligence.²

The line is one clause. What it protects is the hour the tool has spent all summer trying to take.

Almost every vendor pitch that reached a superintendent's inbox this year led, one way or another, with the same offering. Grade faster. Return work sooner. Free the teacher up for other things. Some pitches were hedged carefully. Some were not. The through-line was the same. The tool can carry the grade.

The Orange County clause quietly refuses that offering. The grade is not something the software gets to hand back. It is a decision the adult in the room is still required to make.

Twenty Thousand Voices

Twenty thousand responses to a policy questionnaire is an unusual number for a K-12 district.¹ Most district AI policies this year have been drafted by a small committee, released on a Tuesday, and adopted the following Thursday with two speakers at the microphone. Orange County drafted its policy publicly enough that a fifth-grade parent could weigh in and a middle school teacher could disagree, and both of them did, in enough volume to fill twenty thousand text boxes.

The board member who voted against the final policy, Stephanie Vanos, did not vote against it because it went too far. She voted against it because it did not go far enough on one specific point. It contained no provision for parents to opt their children out entirely.³ That objection tells you something about the room. The disagreements are no longer about whether to have a policy. They are about how much of the process to see.

The Detector Is Not The Instrument

Refusing AI detection software is, unusually, the direction the season is quietly moving. Several large districts have drawn the same line this summer, most of them in draft form.⁴ Orange County is the largest district this month to lock that refusal into approved policy.

There is a reason worth naming. Detection software points at the finished page and tries to reconstruct, from a probability score, whether the words came from a machine. It is answering the wrong question by looking at the wrong artifact. It cannot tell a teacher whether the student paused for a minute on the third paragraph and rewrote it in her own words. It cannot tell her whether the student pasted a suggestion in and then, on second thought, deleted it. Detection reads the surface. What a teacher needs is the record of the hour.

What The Two Clauses Point At Together

Read the grading clause and the detection ban next to each other. They point, from different sides, at the same missing layer.

If the teacher cannot use software to catch the shortcut, and she cannot delegate the grade to the same class of software that produced the shortcut, then the only thing left for her to look at is the process itself. The pause. The revision. The paragraph deleted and rewritten in different words. The moment the student picked up a chatbot suggestion and translated it back into her own language, and the moment she did not.

None of that appears on the final document. All of it is what would let her tell, on a Monday morning in September, what her students actually did over the weekend. At Koan we work on that layer, and this week Orange County drew a line under the fact that a policy alone will not surface it. The instrument still has to be built.

What The Fall Will Actually Ask

The Orange County vote is not, in the end, about which tools to buy. It is about which decisions to keep human. The grade is now, by written policy, one of them. The detector, by written policy, is not.²

What sits between those two lines is the classroom itself, and what a teacher can actually see happening in it before Friday. The policy has drawn the line. Whether Orange County's teachers walk into fall with something to look at, and not just something to enforce, is the question the twenty thousand voices did not answer.

If your district banned AI detectors and required humans to keep the grade, what would your teachers actually look at on Monday morning?