The Conversation, Not the Detector

On Monday, May 18, the Oaklandside published a piece about Oakland Unified School District's draft policy on artificial intelligence.¹ The headline borrowed the same phrase teachers had been using all year. Wild West. After more than a year of classroom-by-classroom improvisation across a district of roughly 35,000 students, OUSD finally has language to govern the technology that has already arrived. The draft, the reporter wrote, takes no stance on whether teachers and students should use AI. It provides guardrails for those who do.¹

The interesting part of the draft is not what it allows or prohibits. The interesting part is a single instruction about what teachers should do when they suspect a student has handed in work that is not theirs. The policy does not tell them to run it through a detector. It tells them to sit with the student. Chat. Gauge their familiarity with the contents of the assignment. Ask them to walk through what they wrote.¹

The Quiet Break

This is a quiet break from where the conversation has lived for two years. Most early district responses treated AI misuse as a forensic problem. The right scan, the right percentage, the right fragment flagged. The trouble was that the detectors did not work and never quite would. A 2023 Stanford study found that more than half of essays written by non-native English speakers on the TOEFL were misclassified as AI-generated, while native eighth-grade writing was identified accurately.² The detectors were not catching cheaters. They were quietly penalizing students whose sentences were shorter and whose vocabulary was narrower.

Oakland's instruction is the opposite. Stop trying to find a watermark in the output. Look at the student and ask whether the writing belongs to them.

It is the right instinct. It is also, in 2026, a very hard one to scale.

The Math of the Conversation

A high school English teacher in Oakland may have 150 students across five sections. The policy is asking her, in effect, to be ready to hold a one-on-one viva any time she has a question about a piece of writing. The dialogue is the right next step. The minutes are not there.

It turns out the detection question and the dialogue question are the same question asked from opposite ends. The detector asks the output: are you human-made? The dialogue asks the student: did you make this? The first is unanswerable. The second is answerable, but only if the teacher walks into the conversation with evidence.

Evidence is the part that has been missing. A teacher walking into a conversation with no record of how the work was made is starting from zero. The student is asked to reconstruct her own process from memory while also being judged on it. This is not a productive setup for either person.

What the Policy Quietly Assumes

Oakland's draft assumes something that most districts have not yet built. That the teacher will be able to know, before the conversation begins, roughly how the work came together. Otherwise the conversation is just a guess against a guess.

What would let a teacher walk in already informed is not a stronger detector. It is the trace of the work itself. Drafts in the order they were written. Pauses where the student stopped to think. The questions a student asked an AI tutor and what they did with the answers. A revision that came twenty minutes after feedback. Small, ordinary signals that, gathered into a record, let a teacher walk in already knowing most of what she needs to know.

This is the bet we have been making at Koan. The workspace captures the process. The conversation with Aidan, the drafts, the revisions, the pauses. When a teacher has a question, she does not start from zero. She starts from a record. The conversation Oakland is asking for becomes a five-minute clarification instead of a thirty-minute interrogation.

Policy Without Infrastructure

By July 1, every public district in Ohio will have an AI policy of its own under a state mandate that took effect last summer.³ Most of them will say something close to what Oakland is saying. Use judgment. Don't rely on detectors. Talk to your students. These are good words. They are also a kind of unfunded mandate. Without a record of the work, a teacher with 150 students cannot have 150 conversations.

Forty-seven percent of educators told the EdWeek Research Center in a survey reported this winter that they think AI will have a negative impact on teaching and learning. Forty-three percent said it would be positive.⁴ The split is not really about the technology. It is about what the classroom around the technology looks like. If the policy is "talk to the student about their work" and there is no record of the work, the policy will quietly fail. If there is a record, the policy might be the most useful thing a district has written in a decade.

The Right Question for the Next Year

Across the country, the policies are converging on roughly the same instinct. Don't trust the detector. Trust the conversation. The next twelve months will not be about whether that instinct is right. It is right. The next twelve months will be about whether the classroom underneath it can hold the instinct up.

If the policy is now "talk to the student about their work," what is the work showing the teacher before the conversation begins?