The Sixth Practice
On May 12, the Michigan Department of Education released its first comprehensive AI guidance for K-12 districts. Six essential practices, calmly listed. Five of them can begin tomorrow. The fourth and the sixth quietly name an infrastructure that does not yet exist in most schools.
On Tuesday, May 12, the Michigan Department of Education released its first comprehensive AI guidance for the state's K-12 districts. A starter guide for districts just beginning the conversation, and a longer framework for those further along.1 The announcement was modest. State Superintendent Dr. Glenn Maleyko offered a sentence that has by now become a kind of standard for these documents. "AI can be a good learning and teaching tool if used properly. But we need to make sure our educators and our students understand AI and know how to use it appropriately."1
The guidance arrived in a state where the work was already well underway. Michigan Virtual, the state's online learning agency, has trained more than 25,000 teachers in the basics of using AI in their classrooms.2 At Williamston Community Schools, just outside Lansing, Superintendent Adam Spina told reporters that his district had been adapting to the technology for months.2 The state's announcement is not, in other words, the beginning of a conversation. It is the moment when a state names what it has been quietly watching.
The substance of the guidance is a list of six essential practices.1
- Keep AI purposeful and safe.
- Protect privacy and integrity.
- Build AI literacy.
- Maintain human oversight.
- Promote equity and accessibility.
- Support transparency and continuous improvement.
Read aloud, they have the calm shape of a list any school could nod along to. It is the kind of list that arrives in a press release and disperses, in implementation, into committees and policy templates. But the list rewards slow reading. Look at the verbs.
Five Things Policy Can Do
The first three practices are things a district can begin tomorrow with a memo. You can declare an AI tool purposeful by choosing it carefully. You can protect privacy by writing a vendor checklist. You can build literacy by funding professional development, of the sort that has already touched 25,000 Michigan teachers.2 The fifth, equity, lives in procurement decisions and access planning. These are real practices, and they are difficult, but they belong to the world of policy. They can be set in motion by a meeting.
The fourth and the sixth are different. "Maintain human oversight" and "support transparency and continuous improvement" do not describe a policy. They describe a relationship between a teacher and a student, mediated by a tool that produces and consumes language at a speed no human can audit by hand. They require a thing most schools do not yet have. They require a record.
What Oversight Needs to See
It is easy to write "maintain human oversight" into a guidance document. It is much harder to say what the human would oversee. A finished essay does not tell you whether the student asked an AI to write it, asked an AI to argue with it, asked an AI to help them understand a single confusing paragraph, or never opened the tab at all. The output is, in the language of evidence, drastically underdetermined by the process. Oversight without a record of the process is judgment without information.
"Continuous improvement" has the same problem in a longer time horizon. To improve, a teacher must be able to look back at last week's lesson and see, with some precision, where students struggled, where they accelerated, where AI helped and where it quietly became a crutch. None of that lives in a grade. It lives in the small, ordinary acts that produced the grade. The pauses. The rewrites. The question a student asked at 9:47 on a Tuesday night, then thought through and answered for themselves.
The Quiet Direction
This is the direction the most useful policy guidance has been pointing, even when the documents themselves do not say it out loud. Six practices, calmly listed, and the last two are asking for an infrastructure that does not yet exist in most schools. Michigan has not built that infrastructure. No state has. The market is currently full of tools that generate student work, and almost empty of tools that make the work of producing it visible.
This is the bet behind Koan. When a student writes inside a workspace that captures their drafts, their pauses, their questions to a tutor, their corrections after feedback, "human oversight" stops being a phrase in a guidance document. It becomes a thing a teacher does on Wednesday morning, with their coffee, scrolling slowly through what a student actually did. "Continuous improvement" becomes the same act, repeated across a class, across a semester, across a year. The record is what makes the practice possible.
Michigan's six practices are the right six. The work ahead is not to argue with them. It is to build the tools that let the fourth and the sixth move out of the documents and into the rooms.
If the policy says "maintain human oversight," and the only thing the teacher can see is the finished work, what exactly is being overseen?
References
Artificial Intelligence Guidance from MDE Helps School Districts Use Emerging Technology
Michigan Department of Education · May 12, 2026
Michigan schools navigating growing use of AI with new state guidance
13abc · May 16, 2026
Sources cited in order of appearance. Click any inline number to jump.