Closer to the Student

On Wednesday, May 20, NPR published a quiet, carefully reported piece with a headline that read like a small surrender. "Overworked and understaffed: Special ed teachers turn to AI for help."¹ The reporting was anchored in a new survey from the nonpartisan Center for Democracy and Technology. Fifty-seven percent of special education teachers polled said they used AI to help develop individualized education programs during the 2024-25 school year. The year before, the number was 39 percent. An eighteen-point jump in a single school year.²

The story is being read this week as an AI story. It is more interesting as a labor story. And more interesting still as a question about what classrooms are for.

The Hours That Used to Cost Something

A special education teacher is, on paper, an educator. In practice, much of her week is documentation. Goals to draft. Progress to log. Data to synthesize for the parents who will sit across from her on Thursday afternoon. The IEP is the legal spine of a student's school experience, and every annual one runs dozens of pages. A teacher with thirty caseload students will rewrite close to a thousand pages of plans in a year.

The survey result is not surprising. It is what overworked people do when handed a tool that can draft on their behalf. Ariana Aboulafia, who led CDT's report and runs its Disability Rights in Technology Policy Project, called the tools "a Band-Aid" for a profession that has been bleeding minutes for a decade.¹ A Band-Aid is what you reach for when no one has fixed the cut.

What the Time Back Could Mean

The most quietly hopeful sentence in the NPR piece came from a researcher describing something almost obvious. "The more face time a student with a disability has with a teacher, that often yields better outcomes for them, both educationally, functionally, just across the board."¹ More minutes next to a student is not a soft virtue. It is a measured one.

Picture what a teacher does with two extra hours on a Tuesday afternoon. She sits next to a sixth grader who has been stuck on the same paragraph for three days and asks him to read it aloud. She notices, for the first time, that he loses the thread at the same comma every time. She makes a note for Thursday's meeting that has nothing to do with the form. It has to do with the comma. That was the work the form was meant to describe. It was just never the form.

The Fifteen Percent

The same CDT brief flags a number that has unsettled disability advocates. Fifteen percent of teachers surveyed said they rely entirely on AI to develop IEPs.² Aboulafia is most worried about this group. AI models, she points out, are pattern-recognition engines. Individualized education programs are, by their statutory definition, the opposite of patterns. They are documents about a single person, and the law that produces them, IDEA, treats individualization as a civil right.¹

Generative tools, trained on the existing corpus of special education writing, also tend toward the language of deficit. "Requires frequent adult prompts." "Struggles to remain on task." Once written into a plan, those phrases follow a student for years. Transition teams in high school read them and quietly recommend more restrictive programs. The pattern, generated in seconds, becomes the destination.³

The real tension is not between teachers who use AI and teachers who do not. It is between teachers who use it to free themselves for the student in front of them and teachers who use it in place of that student. One returns to the human. The other replaces her.

What an IEP Should Be Resting On

Research from teams at the University of Virginia and the University of Central Florida, cited in the NPR piece, suggests AI-assisted IEPs can be as good as or better than ones written by a teacher alone, when the teacher stays in the loop and the AI stays in service of evidence.¹ That qualifier is doing real work. The IEP is only as strong as the record of the student the teacher carries into it.

This is where the conversation gets close to what we have been building at Koan. An IEP, at its best, rests on a record. The drafts a student wrote and rewrote. The questions she asked the tutor and the ones she answered for herself. The pause she took before the comma. When the work is visible, the form has something specific to describe. When the work is invisible, the form drifts toward generic patterns, and the patterns drift toward generic kids.

The Question Underneath

The headline this week is that AI has arrived in special education. The story underneath it is older. We built a profession around seeing individual students. We then built a documentation system that took most of the hours the profession had. Now we have a tool that can do the documentation. The next question is not whether to use it. It is whether the classroom around it is built to use the hours the way the profession originally promised to.

The Band-Aid stays useful for as long as the wound stays open. The interesting work is closing the wound.

If AI can give a special education teacher two hours back on a Tuesday afternoon, what does the rest of her classroom need to look like for those two hours to land on the student?