Program before platform · Niall Highland

Almost every school I visit has already made its first AI decision, and almost every school made it in the same order. Someone — a vendor, a board member, an enthusiastic head of department — arrived with a platform. It looked impressive. It promised to "handle AI." And so the school bought the tool first and then, quietly, began the work of inventing a learning program that would justify the purchase.

I understand exactly how this happens. The platform is concrete and the program is abstract, so the platform feels like progress. But it is the wrong way round, and the schools that will look back on this period without regret are the ones that inverted it: they decided what they wanted learning to look like, and only then chose the tools to serve it.

Key point

The learning program comes first. The tools are chosen to support it — never the other way around.

That sounds obvious written down. It is surprisingly rare in practice, because doing it in the right order means sitting with uncertainty for longer than a purchase order takes to sign. What follows is the sequence I've watched work — not a policy to adopt, but a process to run.

5moves, in order — and none of them is “choose a platform”

3voices the task force needs at minimum: teachers, students, leadership

∞revisions — the guidance is never “done”

01Align on why, before what

Before a single guideline is drafted, the leadership team has to agree on what the school's AI strategy is for. Not the tools, not the wording — the purpose. What are we trying to protect? What are we trying to make possible? A leadership team that hasn't answered those questions out loud will discover, six months later, that it has been building policy on top of three quietly incompatible assumptions.

This is the least glamorous step and the one most often skipped. It is also the one that makes every later step faster.

02Build a task force that actually represents the school

The next move is to form a representative AI task force: teachers from different subjects and divisions, leadership, and — this is the part schools underrate — students. A maths teacher and a drama teacher do not have the same relationship to generative AI, and a policy written by one of them will quietly fail in the other's classroom. Students, meanwhile, know precisely how AI is already being used in your building, which is information you will not get any other way.

The task force's job is not to hand down rules. It is to make sure the questions being asked are the real ones.

03Draft working guidelines — then test them in real classrooms

Now, and only now, you draft an initial set of working guidelines. The word working is doing a lot of labour: these are not the policy. They are a first attempt, deliberately provisional, whose entire purpose is to be tested across different subjects and grade levels so the gaps and unanswered questions surface where you can see them.

The most useful thing a set of guidelines can do at this stage is establish a shared language — a common vocabulary a whole faculty can use to talk about when and how AI may be used. The simplest version I've seen is a traffic-light model. Try it below: sort a handful of ordinary classroom tasks into the zone where each belongs.

A shared language for AI useThe traffic-light model

Sort each task into the zone where it belongs. There are no trick questions — only the conversation a faculty has to have out loud, once, so it never has to have it again.

RedDo it yourselfAmberUse, then discloseGreenUse freely

A student uses AI to brainstorm essay angles, then writes the essay themselves.
A student asks AI to rewrite their paragraph to “sound smarter,” and submits it.
A student uses AI to translate a foreign-language source, quotes it, and notes the tool.
A student pastes the homework prompt into AI and submits the answer verbatim.
A student uses AI to check their math and explain the problems they got wrong.

0 / 5 classified

Interactive Fig. 1The value isn't the verdict on any single task — it's that a whole faculty can reach for the same three words. Shared language is what makes the hard cases quick.

Notice how little the exercise depends on a document. What resolves the awkward cases is not a rule you looked up but a vocabulary you share. That is what you are actually building at this stage — and it is why testing the language across subjects matters more than perfecting the wording.

04Share it widely, and mean it

Working guidelines that live inside the task force are worth very little. The fourth move is to share them with the wider community and actively gather feedback — from teachers who will have to teach inside them, from students who will have to learn inside them, and from parents who are, right now, comparing schools on exactly this question. Active is the operative word. A guideline posted to the staff drive and never mentioned again has not been shared; it has been filed.

The feedback will not be tidy. It will expose disagreements you would rather not have found. That is the feedback doing its job.

05Revise on a schedule — because the ground moves

The final move is not a finish line. It is a commitment to review and revise regularly, because the technology underneath all of this changes on a timescale that no annually-approved policy can keep up with. AI guidance has to be treated as a living document rather than something written once and filed away.

Schools that internalise this stop trying to write the correct policy and start maintaining a current one. That shift — from document to practice — is the whole game.

Why the order matters

	Platform first	Program first
Starting point	Platform firstA tool is bought; a rationale is written to fit it.	Program firstA purpose is agreed; tools are chosen to serve it.
Who writes the rules	Platform firstA vendor's defaults, lightly edited.	Program firstA representative task force, tested in real classrooms.
When the model changes	Platform firstThe policy — and often the tool — has to be rewritten.	Program firstThe living document absorbs it; the program holds.
What you own at the end	Platform firstA subscription.	Program firstA shared language your faculty can actually use.

The comparison isn't quite fair, and I know it — plenty of schools that bought a platform first are doing thoughtful work. But the ones doing the most durable work are, almost without exception, the ones who could tell me what they wanted learning to look like before they could tell me what they had bought.

The starter shelf

Four things worth reading before you draft anything

If you're beginning this work, these are the resources I keep returning colleagues to:

Co-Intelligence: Living and Working with AI by Ethan Mollick — the best accessible starting point I've found for leaders trying to understand both the potential and the implications of generative AI.
One Useful Thing — Mollick's blog, and one of the best sources for keeping up with developments, particularly in education and the workplace.
Pineapple Charts — a simple model for helping teachers observe and share what they're trying in their classrooms. Something similar could surface the good AI practice already emerging quietly in your building.
A shared-language framework — such as the traffic-light model above. Whatever form it takes, agree on the words before you argue about the cases.

None of this requires you to know what the technology will look like in three years. It requires you to know what you want learning to look like, and to keep the guidance alive as everything underneath it moves. Get the order right, and the platform — whichever one you eventually choose — becomes the easy decision it should always have been.