The leader's real job with AI

You are being asked about AI from two directions at once. The board wants to know what the school is doing about it. The faculty wants to know whether they are allowed to use it, and half of them are already using it without asking. Somewhere in between, a parent has emailed about a cheating incident and a vendor has offered you a pilot.

The instinct is to go learn the technology first, so you can answer from expertise. That instinct is wrong, and it will cost you a semester. You are not going to out-learn the teacher in your building who has been reading about this every night since spring. You do not have to. Your job here is the same job you already have: set the frame, model the habit, and keep the decisions that belong to a person.

This lesson is about separating those three so the rest of the course has somewhere to land.

Before you start

• Nothing to install, nothing to open. This lesson is thinking and one page of writing. You pick a tool in lesson 2; picking one earlier is how people spend weeks comparing products instead of deciding anything.
• Last week's calendar, open in front of you. Not a typical week — last week, with the interruptions in it. One exercise runs on it directly.
• A blank page nobody else will see. What you write here is deliberately bad, and you rewrite it in lesson 6.
• Thirty-five minutes, about fifteen of which is the exercise at the end.
• One thing to know now: whatever your district has already agreed to — a contract, a data privacy agreement, a superintendent's memo — sits above anything you decide in this course. Find out what it says before you announce anything. If nobody in your building knows, that is your first email.

If your district has already published AI guidance, print it. You are not contradicting it; you are translating it into something a teacher can act on at 10pm.

Four decisions that are actually yours

Strip away the noise and there are four, and only four, real decisions in front of you this year.

What is allowed. Which uses are fine without asking anyone. For most buildings the honest answer is broader than staff assume and narrower than the early adopters would like.

What is required. Almost nothing should be, at first. But there will be one or two places where inconsistency hurts students more than caution does — usually how AI use is disclosed on student work, and what a teacher says when a kid asks.

What is forbidden. Short, absolute, and about data more than about pedagogy. Student records, health information, discipline files, anything identifiable.

Who decides the rest. The teacher, the department, the principal, the district. Naming this is most of the work. Teachers are not paralyzed because the policy is missing; they are paralyzed because they do not know whether the call is theirs.

Everything else — which tool, which training, which pilot — is downstream of those four and much less important than it feels.

Here is the test of whether you have decided them: each has to survive being said out loud, in a sentence, to a skeptical adult who will repeat it. "Teachers may use it for their own work, never with student data, and each teacher decides whether students may use it on a given assignment" is a decision. "We're exploring responsible AI integration" is a press release. If you cannot say yours in the shape of the first, you are not done deciding, and a faculty meeting will expose that fast.

The three-pile sort

Take last week. Sort what you did into three piles.

Production. The output is the point and you can check it fast. The family letter. The staff memo. The board summary. The grant narrative. The agenda. The forty-page document you needed the gist of. The same message rewritten for three audiences.

Preparation. The value is in your thinking before a moment only you can be in. The difficult conversation with a teacher. The meeting where you have to say what you actually believe about a proposal. The parent who is coming in angry and will be listened to.

Presence. The job itself. The conversation. The decision and owning it. Standing in the hallway at 7:40. Telling a family something hard. Being the adult who noticed.

AI belongs in the first pile fully, in the second as a questioner, and in the third not at all.

The sort is not an intellectual exercise. It is a budget. Most leaders find that the production pile is where their evenings go and the presence pile is what gets cut when the week goes sideways — which is backwards, because the production pile is the only one that can be handed over. When you circle the production items, you are not listing AI use cases. You are listing the hours you are trying to buy back, and lesson 3 is how you spend them.

One caution on the middle pile. Preparation is where a tool is genuinely useful and also where it is easiest to let it do too much. Use it to generate the questions you have not thought of and the objections you will face. Do not use it to decide what you think. That difference shows up in the room, fast.

The test, when you are not sure

For anything: if this came back wrong, would I know, and would it matter?

• I would know and it would not matter much — hand it over, review it, move on.
• I would know and it would matter — take the draft, keep the decision, read every line.
• I would not know — do not hand it over. You cannot review what you cannot judge.
• I would know, but the point is that I showed up — keep it, entirely.

That last line is where leaders get burned. A condolence note, a note to the teacher who just got through a brutal week, the message after something went wrong at your school — a machine writes all of them competently, people can usually tell, and what they learn is that you did not have four minutes. The cost is never the note. It is the next conversation.

The third line should slow you down wherever you are least expert. A legal summary, a special-education procedural question, a grant's eligibility rules — these come back fluent, confident, and sometimes wrong in ways you are not equipped to catch. Fluency is not accuracy, and from the outside they look identical. There, use the tool to generate the question you take to the person who does know, not the answer you act on.

What you are modeling, whether you mean to or not

Your staff is watching what you do far more closely than what you write. If the first AI thing you do in public is an all-staff email that reads like a brochure, you have taught the building that AI writing is the thing you send when you do not care. If the first thing you do is say, in a meeting, "I used it to get the first draft of this letter to families, then I rewrote the second half because it was too formal, and here is what it got wrong" — you have taught them a usable habit and given them permission in the same breath.

Say the quiet part out loud early: I use it. Here is where. Here is where I do not. That single sentence does more than a policy document.

The corollary is uncomfortable: you cannot hold teachers to a data line you are not visibly holding yourself. If you paste a student's behavior history into a chatbot to get advice on a conference, you have written the real policy, and it is not the one on the wall.

What you must keep

Three things never move, no matter how good the tools get.

Judgment about a specific child. The tool has never met them. Any recommendation it makes about one student is a guess wearing the costume of advice.

Anything you would have to defend. A discipline decision, a personnel decision, a placement. If you could not explain the reasoning without referring to what a tool said, you do not have reasoning yet.

Accountability. "The system flagged it" is not an answer to a parent, a board member, or a hearing. The name on the decision is yours. That does not change, and you should not want it to.

A worked example

The principal and the school here are invented; the output is illustrative.

Marisol is in their second year as principal of a middle school. The district has a one-paragraph AI statement saying staff should use approved tools "responsibly," which has answered nothing. Two teachers asked in the same week whether students may use AI on a research paper, and Marisol gave them different answers, which they noticed immediately.

So they write the four decisions badly, in one line each, on a legal pad. Then they run this — not to get the answers, but to find out where their own answers fall apart:

I am the principal of a middle school. Below are four decisions I have
made about AI use in my building, written quickly and not polished.
Do not rewrite them and do not give me better versions.

1. Allowed: teachers can use it for their own work — planning,
   drafting, rewriting, summarizing long documents.
2. Required: students say on the assignment when they used it; any
   adult asked "am I allowed to use this" says the same sentence.
3. Forbidden: nothing with a student's name, records, IEP or 504
   content, health information, or discipline detail goes into it.
4. Who decides the rest: the teacher, per assignment.

Do three things. First, tell me which of these four a teacher could
follow on a hard Tuesday without asking me, and which one they could
not. Second, name the situations these four do not cover at all.
Third, for each gap, tell me who in a school district would normally
own that decision — me, the department, the district, or counsel.

What came back:

Decision 3 is followable without you: it is a list of nouns, and a teacher can check their own behavior against it. Decision 1 is nearly followable, but "their own work" is carrying a lot — a teacher writing report-card comments is drafting their own work and also producing something that lands in a student's record. Decision 4 is not followable as written: "per assignment" tells a teacher they own the call but not what the floor is, so a cautious teacher still emails you and an aggressive one does not.

Gaps: students using AI outside of school on school work; paraprofessionals and front-office staff, who are not "teachers"; substitutes; a teacher using a tool the district has not reviewed; what happens the first time a student does not disclose.

Marisol changed three things. They split decision 1 so anything landing in a permanent record gets a named human decision — the report-card case was real and they had missed it. They rewrote decision 4 to say the teacher decides, inside a floor the school sets, and the floor is on the one page, which is the sentence that ended the two-teachers-two-answers problem. And they wrote "staff" everywhere they had written "teachers," because the front office handles more sensitive material in a week than most classrooms do.

What they did not do is treat the list of gaps as a to-do list. Most get answered in lesson 6, on one page, and some belong to the district.

Do this now

Fifteen minutes, on your own building, before you read lesson 2.

1. Write the four decisions — allowed, required, forbidden, who decides the rest — in one line each, badly. Do not polish, do not send. Speed is the point: you are finding out what you already believe.
2. Read each one out loud as if to a skeptical teacher. Cross out any that come out as "we are exploring" rather than as a decision.
3. Open last week's calendar and sort every entry into production, preparation, or presence. Circle the production items.
4. Count the circled items and write the number at the top of the page. That number is the case for the rest of this course.
5. Write one sentence naming the thing on the presence list that got cut when the week went sideways.

You are done when one page in front of you has four decisions in four lines, a circled production list with a count at the top, and that one sentence. Keep the page — lesson 6 starts by making you read it again.

If it goes wrong

• You cannot write the "forbidden" line because you do not know what the district already forbids. Write what you believe and put a question mark next to it, then email whoever owns the district's data agreements. The page is provisional either way; not knowing is not a reason to stall.
• All four come out as "it depends." You are writing for every possible case instead of for the next month. Rewrite each for the specific thing that happened in your building last week. Specific decisions generalize later; general ones never apply.
• The production pile swallows everything and the presence pile is nearly empty. That is a finding, not a failed exercise. It usually means you sorted by the calendar rather than by what you did — the hallway, the doorway conversation, the walk to the bus are presence work and never get scheduled. Add them by hand.
• You start drafting the policy document instead of the four lines. Stop. The document is lesson 6, and it will be worse written now, before you have made the two decisions it has to encode.
• Someone asks you for the policy while you are still on this lesson. Give them the interim answer out loud: teachers may use it for their own work, nothing identifiable goes in, and the full page is coming by a date you name. An interim answer with a date attached stops the guessing. Silence does not.

Resources

• TeachAI: AI Guidance for Schools Toolkit — read it when you need to see what other districts actually put on paper.
• ISTE+ASCD: Artificial Intelligence in Education — use it for vocabulary your teachers already recognize from other professional learning.
• CoSN: K-12 Education Technology Leadership Association — for the technical, district-level questions: data, infrastructure, procurement.
• Protecting Student Privacy (US Dept. of Education, PTAC) — open this before you write your "forbidden" line, so it matches the law you already operate under.
• AASA, The School Superintendents Association — what the conversation looks like one level above you, where your one page has to fit.