What The Children Told Her

On July 7, Chalkbeat's Newark bureau published a small piece with a very large question underneath.¹ In May, Randi Weingarten, president of the American Federation of Teachers, had visited Park Elementary School in Newark to see Khanmigo, Khan Academy's AI-powered tutor, in the district's flagship implementation. Newark's own press release described a district-leading showcase.² Weingarten's account, offered publicly for the first time this month, was quieter. She had gone in hoping the visit would change her mind. It confirmed her fears.¹

The disagreement between the two accounts is not what people usually mean by disagreement. The district was not staging a lie. Neither was Weingarten. Newark has built a real partnership with Khan Academy. Fourteen of its schools now use Khanmigo, mostly in grades three through eight. Last year the district became the first collective recipient of the Khan Academy Award for Excellence and Innovation.³ Weingarten met the teachers who lead the program. Superintendent Roger León walked her through it. All of that is real.

The other real thing is what the children said when she asked them. Weingarten reported that when she spoke with students about Khanmigo, they told her the tool reinforced skills they had already learned, and did not teach them anything new.¹ It was in a corner of the room, where a teacher had pulled a few children aside for a small hands-on science experiment, that their faces changed.

What The Adults Could See

Newark's visit showed everything a visit can show. A district's implementation strategy. A superintendent's vision. Teachers who have been trained. A well-managed vendor partnership. Award-winning outcomes on the metrics that districts and foundations track. What a walk-through cannot show, structurally, is the smaller and older question. What was actually happening in a specific child's mind at a specific minute of a specific Tuesday.

Weingarten's answer to that question is why, the day after her Newark trip, she went further than any major union leader had gone. On May 27, at the National Press Club, she rolled out a ten-point plan called Devices Down, Eyes Up, Hands-On.⁴ It called for an end to student-facing AI in elementary schools. She was careful to say she was not calling for a Chromebook bonfire.⁵ She was calling for balance. What she meant by balance was what she had seen in the room. The child in front of the screen was not lighting up the way the child at the lab bench was.

The Gap Between Two Accountings

The gap between Newark's press release and Weingarten's notes is not a fight over Khanmigo. It is a symptom of a much larger absence in almost every school in the country. Districts have implementation data. Vendors have engagement data. State tests exist. Nowhere in any of that is the single most important artifact of a lesson, which is the record of what a child was actually doing, from moment to moment, while she was allegedly learning. Whether she paused. Whether she reread. Whether she tried an answer and threw it out. Whether she asked the tool a real question, or waited for it to keep going.

If either side of this argument had access to that record, the conversation would look different. The district would not have to describe its program in the language of excellence and innovation. It could point to a fifth grader on a Wednesday morning and say, here is what she did. Weingarten would not have to rely on a fleeting classroom conversation to know whether the tool was reinforcing skills, introducing them, or quietly replacing the moments when a child was supposed to be building them for herself.

What The Room Would Have Shown

That is the work we spend our days on at Koan. Not another AI tutor to sell to a district, but the ordinary quiet layer underneath the screen where the process of a student's thinking leaves a trace worth reading. The pauses. The half-drafts. The moments of choosing to keep going on her own, and the moments of typing the model's sentence in as if it were hers. If Weingarten's Newark visit had ended not with a hallway conversation with a handful of children, but with a look at the record of what those children had done for the week before she arrived, both accounts might have converged. The country might have moved from an argument about AI in schools to a decision about it.

Newark will keep expanding Khanmigo. AFT will keep pushing its ten-point plan. Every district in the country will decide again this fall whether to trust the demo or the doubts. What almost none of them will have, when they decide, is a way to see on an ordinary October afternoon what a particular child in a particular room actually did with the tool. Until that changes, both parties are guessing. The demo says one thing. The child says another. The tool, having no view of either, keeps going.

If the room a visitor sees is not the room the child lives in, whose account should the district trust when it decides what to keep buying next fall?