Skill Before Shortcut
Updated: Nov 2
AI will keep getting more powerful. The decisions about what stays human cannot wait.
An article from the Build series. Part Two of two.
In Part One, Automate the Mundane to Protect the Human, I made the case for deciding what is sacred before you decide what to automate, and I shared the five guardrails I run my own practice by. Guardrails work for one person. This piece is about building that decision into an institution, so it holds when no single person is in the room.

I will also do something Part One did not: answer the hard questions directly, and say what I think a school should actually do.
One note on language before I start. I have worked in independent schools and in the private sector, and the work is more alike than either side tends to think. When I write "head of school," read president or CEO if that is your seat. When I write "school," read company, firm, or organization. The board is the same table in both worlds. Wherever you sit, find yourself in this.
The Window for Waiting Has Closed
There is a third response to AI besides fear and hype, and it is the most common one in schools: waiting.
The numbers say that window has closed. Stanford's 2026 AI Index reports that about four in five U.S. high school and college students now use AI for schoolwork, while only about half of middle and high schools have any AI policy. Among teachers in schools that do have one, just 6 percent say it is clear. More than thirty states have issued AI guidance for schools, and state boards are moving from guidance toward monitoring. The next wave of tools, AI agents that carry out multi-step work on their own, is already arriving in school business offices.
A school that has not written a policy has not avoided a decision. It has made one by default.
That default gets handed to whoever happens to be in the room. And every year the tools grow more powerful, the cost of having no guardrails grows with them.
What is at stake is not abstract. Families choose independent schools for human contact, and they notice when it thins. A student who uses AI to skip the hard part of an assignment has skipped the reason the assignment existed. Time freed by automation can become growth or cuts, and that is a leadership choice. A single careless upload of family data can cost more trust than a year of efficiency can buy back.
There is an honest environmental cost, too. The data centers that run AI use large amounts of water, mostly through the power plants that supply their electricity. A Harvard study released this summer estimated that 472 large U.S. data centers account for roughly 300 billion liters a year, with the sharpest strain in already dry regions. The answer is not to refuse the tool or to rush it. It is to learn it well enough to govern it.
When Should AI Enter the Room?
I coach youth basketball, and every season starts the same way. You do not hand a young player the playbook before they can dribble with their head up and make a clean pass. Plays only work when the fundamentals are automatic. Hand a team the plays too early and they run the shapes on the floor without understanding the game.
AI is a playbook. It is powerful in the hands of someone who already knows the fundamentals, and hollow in the hands of someone who does not.
Some institutions are already drawing the lines. The most widely used AI chatbots require users to be at least 13, with parental permission for teenagers. Seattle Public Schools restricts several of the most common generative AI tools for students in prekindergarten through fifth grade, because the safety filters do not yet work well enough for the youngest learners. Lakeside, which serves grades five through twelve, asks students to get a teacher's permission before using AI on any assignment.
The principle underneath those choices is simple enough for any school to adopt: skill before shortcut. A student should be able to do the thinking before a machine is allowed to do it for them.
Schools That Have Already Written It Down
Schools at every level have stopped waiting and started writing. The best of them wrote down a purpose, not only a rule.
In the early grades, Gwinnett County Public Schools in Georgia, one of the largest districts in the country, built what its director of AI and computer science calls "an intentional K-12 learning progression." Its youngest students are not handed AI tools. They learn what AI is, so it is less mysterious by the time they are ready to use it.
In high school, Phillips Academy Andover treats AI as a learning aid, not a replacement for thinking: a student might generate a paragraph or a block of code with AI, then have to analyze and explain it alone. The school puts the purpose plainly: "the irreplaceable human element will always be at the core of an Andover education." Phillips Exeter is leaning on Harkness, its discussion method since the 1930s, to build the critical thinking students need around AI.
In college, the purpose shifts from protecting the fundamentals to building fluency. Ohio State announced in 2025 that AI literacy would become a baseline expectation for every undergraduate, regardless of major, beginning with the Class of 2029.
Put them side by side and a ladder appears. Elementary schools teach what AI is. Middle and high schools teach when to use it and when not to. Colleges teach how to use it well in a field. That is skill before shortcut, written across a whole education.
What the Field's Own Data Says
That ladder is not my invention, and it is not the work of a few schools either. In September 2026, the National Association of Independent Schools (NAIS) published AI in Independent Schools: Policy Formed, Practice Forming, built on surveys of more than 800 school leaders and teachers. School by school, without coordinating, leaders described nearly the same sequence: little or no direct student use in the early grades, a structured start around fourth or fifth grade, middle school as the hardest ground, and upper school as the place where real use begins. One division head called it being "second through the door." I would call it skill before shortcut.
The report's title says the rest. Policy has formed. Practice is still forming.

Read it as a builder, and three gaps stand out.
The students have rules. The adults do not. The share of teachers reporting no student AI policy fell from 29 percent to 5 percent in a single year. Yet a quarter of division heads said their school had no AI policy for faculty and staff, even though the adults are the heaviest users in the building: 63 percent of division heads use AI weekly or daily, while only 6 percent of teachers use it daily in the classroom.
Clarity is what saves the time. Teachers at schools with clear frameworks described AI as a time-saver. Where guidance was thin, the work of deciding what counted as responsible fell to each teacher alone. A policy is not paperwork that slows the tool down. It is what lets the tool save time.
The families are waiting. Fifty-six percent of division heads said their school was considering AI training for parents. Eight percent were offering it. Much of a student's AI use happens at home, and most families are being left to work it out on their own.
None of those are technology gaps. They are design gaps, and design is what this series is about.
The Same Pattern Outside Schools
Schools and companies look exactly alike here.
Gallup found in 2025 that leaders were twice as likely as individual contributors to use AI frequently, yet only 22 percent of U.S. employees said their organization had communicated a clear plan for it.
Boston Consulting Group's global AI at Work 2025 survey found frontline use stalled at 51 percent, only 36 percent of workers feeling adequately trained, and more than half willing to use AI tools even if they were not authorized.
A KPMG and University of Melbourne study of more than 48,000 people in 47 countries found that only 40 percent of workplaces had any AI policy or guidance, and that 57 percent of employees had hidden their use of AI at work.
Put those beside the NAIS numbers and the same shape appears in every sector. The people at the top adopt first. The people closest to the work are left to figure it out. And where no one writes the rules down, people write their own, quietly. That is the default I described at the start of this piece, now with numbers on it.
Your people are already using AI. The only question is whether they are doing it inside a policy or around one.
The Growth AI Can Take Away
Here is why the sequence matters so much. The research on what happens when AI arrives too early is still young, but it is already pointing in one direction.
A 2025 study by Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers and found that the more people trusted AI, the less critical thinking they applied to its answers. The people least confident in their own skills were the most likely to accept what the machine produced.
The youngest workers are already feeling it. Stanford economists Erik Brynjolfsson and Bharat Chandar, using payroll data from millions of American workers, found that employment for 22- to 25-year-olds in the jobs most exposed to AI fell by a double-digit percentage after late 2022, while employment for older workers in the same jobs grew. The entry-level job is where people learn judgment by doing the unglamorous work. If AI takes that work, it takes the classroom with it.
A seventh grader and a 23-year-old analyst are facing the same risk. Both are still building the skills that AI makes it tempting to skip.
The struggle is not a flaw in how people learn. It is how people learn.
For young learners and young workers, protecting that struggle is the most human thing a school or an employer can do. And protection that depends on one thoughtful teacher, or one careful manager, is not protection. It is luck.
Why This Is a Build Question
This series is about design. The question underneath every piece in it is the same: how do you build something that keeps creating value after you are no longer the one holding it together?
Right now, in most schools, the rules for AI live in individual heads. One teacher bans it. Another builds assignments around it. The business office uses it and hopes no one asks. The admissions office wonders whether a family's essay was written by a person. Each of those adults may have good judgment. But judgment that lives in one person leaves when that person leaves.
A guardrail is personal. A policy is architecture.
The Questions, and My Answers
These are the questions I would ask any leadership team about AI. Most teams answer them slowly. Here is how I would answer them.
Does your organization have a standing AI policy? It should, and it should fit on one page. Not a paragraph buried in the student handbook, but a written policy the board has approved and the people it covers actually know.
If one is drafted, who is holding the pen? The head's office, with a working group doing the drafting. A policy written by one office for everyone else will miss how students actually use these tools and what families actually fear.
Who owns the use of AI in your institution? The head of school, or the chief executive in any organization. Not IT. Not a committee. Technology teams carry it out, and committees advise on it, but ownership belongs to the person accountable for the mission, because AI policy is mission policy.
Who should be your thought partners? Every seat: students, faculty, staff, families, trustees, and at least one outside voice. Lawrenceville's three councils, which I described in Part One, are the model.
Who holds the final decision? Two answers, and both matter. The board approves the policy and owns its oversight, the way it oversees financial risk. The head makes the day-to-day calls inside it. If everyone in the building cannot name both, the policy is not finished.
The Blueprint: Six Steps from Guardrails to Policy
Here is how I would build it. Each step names who leads it.
A step with no owner is a step that does not happen.
1. Inventory what is already happening. Led by the business office and operations. Before writing a rule, find out where AI is already in use: classrooms, admissions, advancement, the business office, human resources. NBOA has published an inventory tool for exactly this. You cannot govern what you have not seen. Fix any broken process you find before you automate it, which is the lesson Pine Crest taught in Part One.
2. Name what stays human. Led by the head, with the working group. Decide in writing what must stay human: the advising relationship, the admissions decision, the feedback that builds a student's skill, the conversation with a family who is unsure. In admissions and family relations, that means automating the scheduling, the reminders, and the first drafts, and protecting the tour, the phone call, and the moments families remember. Everything else gets measured against this list.
Admissions offices are already showing what it looks like when structure does the work instead of suspicion. At St. Andrew's Episcopal School in Potomac, Maryland, applicants now complete an on-site writing sample after their interview, a step its director of enrollment management called "a way to realign things with our core purpose." In New York City, Beacon High School moved its admissions essays onto campus for the same reason. Neither school bans AI. Both design around it. The protection lives in the process, not in a detector.
3. Set the sequence by age and role. Led by academic leadership. This is where skill before shortcut becomes policy. Here is the sequence I would recommend as a starting point:
Prekindergarten through grade 5: Adults may use AI to prepare their work. Students do not use generative AI tools directly. The fundamentals come first.
Grades 6 through 8: Students use AI only with a teacher's permission, for named tasks, and never for work designed to build the skill itself. Teachers model how to question a tool rather than obey it.
Grades 9 through 12: Students learn AI literacy explicitly: how to check a tool, cite it, and decide when not to use it. Use is allowed where a teacher permits it, and it is always disclosed.
Adults and operations: Data rules come first. No family or student information goes into a public tool. Every AI-drafted message is read by a person before it is sent.
New employees: Protect the entry-level work where people learn judgment. Let them do it by hand first, then teach them to use AI on it.
At home, families can hold the same line. Let a child write the paragraph, solve the problem, and sit with not knowing before a tool does it for them. Then, when they are ready, teach them to use AI as a coach that asks questions rather than a machine that hands over answers.
4. Write it on one page. Led by the working group, approved by the head. A policy people actually follow is short. Mine would have six parts: what we will keep human, who owns this policy, how each group may use AI (students by grade band, faculty, staff, and offices), what data may never go into a tool, how use is disclosed, and when the policy will next be reviewed. The best guardrails live inside the tools, too. Lakeside's in-house AI, which I described in Part One, does this for students.
5. Give the board its governing role. Led by the board chair. AI is a governance question, not only a technology question. The board should see the inventory, approve the one-page policy, and put AI on its calendar the way it treats financial risk: regular oversight, not a one-time vote.
6. Set a review date. Led by the head, reported to the board. The tools will change within a year, and your policy cannot stand still. And when a tool is retired, its data does not retire with it. Dr. Devan Ray Donaldson, who studies AI data stewardship at Indiana University, frames the question plainly: what happens to the data when the AI system is gone? The prompts, logs, and student and family information inside a tool are records, and someone has to own them (Donaldson, 2026). Review the policy at least annually. Treat each review as a chance to move a task from the human column to the automated one, or back again, and to decide what the school will keep, move, or delete when a tool ends.
What I Finally Understood
Part One was about my own practice: let the machine handle the matching, and keep the meaning human. Writing this piece, I came to see that the harder work is not personal at all.
An individual can hold a line with good judgment. An institution can only hold it with architecture: a list of what stays human, a sequence that protects the struggle, an owner, a board that checks, and a date to look again. Without those, every school is one retirement away from forgetting what it decided.
Part One ended by asking whether this is Skynet. This piece ends with a smaller test that matters more: name the person in your building who owns the answer, and point to where it is written down.
Skill before shortcut. Policy before panic. Then let the machine do the mundane, so the people can do the rest.
Read next: Directions, Not Stages: A Compass for Boards and Associations, the four questions every board should ask about the people it serves.
If This Sounds Familiar
If your school is using AI faster than it has decided what AI is for, you are not behind; you are at the decision point. I would welcome a conversation about where the human work lives in your institution, and about building the one-page policy, the grade-band sequence, and the governance that will protect it.
References
Boston Consulting Group. (2025, June 26). Companies must go beyond AI adoption to realize its full potential [Press release on AI at work 2025: Momentum builds, but gaps remain]. https://www.bcg.com/press/26june2025-beyond-ai-adoption-full-potential
Brynjolfsson, E., Chandar, B., & Chen, R. (2025, August). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Stanford Digital Economy Lab.
Chalkbeat New York. (2025, September 30). AI, tutors, parents: Why this NYC school is forcing students to write admissions essays in person. Chalkbeat. https://www.chalkbeat.org/newyork/2025/09/30/beacon-high-school-admissions-requires-in-person-essay-to-combat-ai-tutor
Depelteau, J. (2026, September 8). The admissions step that revealed more. Independent Ideas, National Association of Independent Schools. https://www.nais.org/resource-center/independent-ideas/the-admissions-step-that-revealed-more
Donaldson, D. R. (2026, October 16). Trustworthy information: Building bridges across archives, biomedical informatics, and artificial intelligence [Research symposium presentation]. Information School Research Symposium, University of Washington, Seattle, WA, United States.
Gillespie, N., Lockey, S., et al. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. The University of Melbourne and KPMG. https://www.mbs.edu/2025-impact-report/impact-stories/Global-study-reveals-trust-of-AI-remains-a-critical-challenge
GovTech. (2023, May 15). How one district has learned to build AI literacy. Government Technology. https://www.govtech.com/education/k-12/how-one-district-has-learned-to-build-ai-literacy
Guidi, G., & Dominici, F. (2026). The hidden water geography of U.S. hyperscale data centers in the AI era [Preprint]. arXiv. https://arxiv.org/abs/2607.02531
Holloway, S. (2024, November 1). The age-appropriate application of AI in instruction. AASA, The School Superintendents Association. https://aasa.org/resources/resource/the-age-appropriate-application-of-ai-in-instruction
Lakeside School. (2025). Family handbook 2025-2026: Use of generative AI; Policymaking.
Lee, H.-P., et al. (2025). The impact of generative AI on critical thinking. Microsoft Research and Carnegie Mellon University.
Lee, J. E. (2026, October 12). Automate the mundane to protect the human. Bridging Legacies Across Campuses. https://www.bridginglegaciesacrosscampuses.org/post/automate-the-mundane-to-protect-the-human
National Association of State Boards of Education. (n.d.). States take next steps on governing AI use in schools.
National Business Officers Association. (n.d.). Governing AI integration: An inventory tool for leaders. Net Assets.
National Business Officers Association. (2026, August 31). Your AI questions, answered: Insights from the Byte-Size AI series. Net Assets.
Pendell, R. (2025, June 15). AI use at work has nearly doubled in two years. Gallup. https://www.gallup.com/workplace/691643/work-ai-adoption-takes-hold-employees-leaders.aspx
Phillips Academy Andover. (2024, November 12). AI in the classroom and beyond. https://www.andover.edu/news/2024/ai-in-the-classroom-and-beyond
Phillips Exeter Academy. (n.d.-a). Harkness. https://exeter.edu/node/16
Phillips Exeter Academy. (n.d.-b). Teaching in the age of AI. https://exeter.edu/teaching-in-the-age-of-ai/
Salata Institute for Climate and Sustainability. (2026, July 13). Study maps water footprint of AI data centers. Harvard University. https://salatainstitute.harvard.edu/study-maps-water-footprint-of-ai-data-centers/
Seattle Public Schools. (n.d.). Artificial intelligence: Guiding principles.
Shields, J., & Greene, N. (2026, April 21). Seizing the AI opportunity. Net Assets.
Stanford Institute for Human-Centered Artificial Intelligence. (2026). AI index report 2026. Stanford University. https://hai.stanford.edu/ai-index/2026-ai-index-report
Stringer, G., & Foley, B. (2026, September). AI in independent schools: Policy formed, practice forming [Research report]. National Association of Independent Schools.
The Lawrenceville School. (2026). AI at Lawrenceville.
The Ohio State University College of Engineering. (2025, June). Ohio State launches bold AI Fluency initiative to redefine learning and innovation. https://engineering.osu.edu/news/2025/06/ohio-state-launches-bold-ai-fluency-initiative-redefine-learning-and-innovation





Comments