Why Smart Schools Are Ditching EdTech Subscriptions and Buying an API Key Instead

There is a conversation happening in forward-thinking schools right now. It is not loud yet, but it is spreading. And it goes something like this:

"We are paying for six different AI tools. We did a DPIA on three of them last term. None of them do exactly what we need. And every single one of them is just Claude or GPT with a logo on top."

That last sentence is the one that changes everything.

Every AI EdTech product is a wrapper. You are paying for the wrapping.

Here is something the EdTech industry would rather you did not think too hard about: virtually every AI-powered school tool on the market, the marking assistants, the lesson planners, the feedback generators, the report writers, is built on top of the same handful of foundation models. Claude. GPT. Gemini. That is it. There are maybe four engines running the entire EdTech AI market.

The vendors access those engines via an API. They build a user interface around them. They add their branding. And then they sell it back to you at a per-seat subscription rate that includes a healthy margin for the wrapper.

You are not paying for the AI. You are paying for someone else's interface to the AI.

Now ask yourself: what if your school accessed that same AI directly?

The maths are not subtle

Let us make this concrete. A mid-tier AI model like Claude Sonnet costs around $3 per million input tokens via the API. A million tokens is roughly 750,000 words: more text than most schools will process in a week of staff use. With prompt caching, a feature that reduces cost on repeated queries by up to 90%, and batch processing, which cuts costs by another 50% for non-urgent tasks, the real-world cost of running a school's AI workload directly through an API is genuinely low. Often measured in tens of dollars per month, not hundreds.

Compare that to per-seat EdTech subscriptions. A tool at £10 per teacher per month across a staff of 80 is nearly £10,000 a year. For a wrapper.

The school that owns its API key and builds its own tools is not just saving money. It is spending that money on something it actually controls.

The compliance argument is the one that should keep SLT awake

Data privacy is where this conversation gets urgent, especially for international schools, British schools abroad, and any school operating under UK or European frameworks.

In England, the DfE now requires schools to conduct a Data Protection Impact Assessment for every AI tool they plan to use. The Data (Use and Access) Act has further tightened obligations around how student data is managed, stored, and transferred. And KCSIE 2025 explicitly names generative AI as falling within the scope of statutory online safety obligations.

Here is the problem with buying an EdTech product under these frameworks: the compliance burden does not transfer to the vendor. You still have to do the DPIA. You still have to understand where student data is going. You still have to assess whether data is being transferred outside the UK, processed by third parties, or used to train models. Many popular EdTech AI tools are American products with data stored in US data centres, which means every UK school using them is carrying a compliance risk they may not have fully assessed.

Now consider what happens when you build your own tool on a direct API agreement. You control the data processing agreement. You control what data is sent. You control where it is stored. You choose a model whose terms you have read. The compliance conversation becomes one you are leading, not one you are inheriting from a vendor in San Francisco.

This is not a theoretical concern. The EU AI Act, which comes into full force in August 2026, classifies AI systems that significantly affect educational outcomes as high-risk. EdTech vendors selling into Europe face stringent obligations, and many smaller startups will simply not survive the compliance burden. Schools that have built dependency on those products will be left stranded. Schools that own their own API access will not even notice.

It is not just a UK and Europe problem

The same structural shift is playing out globally, and in some cases it is moving faster.

The UAE Ministry of Education made AI a mandatory curriculum subject from kindergarten through Grade 12 in the 2025 to 2026 academic year, with nearly one million students affected. The goal is not just to use AI tools: it is for students to design their own AI systems, understand bias and algorithms, and practise prompt engineering with real-world scenarios. The UAE is not buying an EdTech subscription for this. It is building infrastructure.

Singapore's Ministry of Education has gone further still. Rather than purchasing AI EdTech products, they built their own tools on top of foundation models and deployed them through the national Student Learning Space platform. The tools, including a Socratic learning assistant and a personalised adaptive learning system, are owned by the government, governed by the government, and aligned precisely to Singapore's curriculum. The model is API-first at national scale.

Australia endorsed a National Framework for Generative AI in Schools focused on curriculum-aligned deployment, not product adoption. The explicit framing is that schools should deploy AI in ways that serve their specific curriculum: which is exactly what a bespoke API-built tool does, and what an off-the-shelf EdTech product categorically cannot.

UNESCO's own research found that 84% of countries lack comprehensive legal frameworks to protect student data, and that the EdTech industry has a documented history of over-collecting and misusing children's data in exactly those unregulated environments. The conclusion of their 2025 report is clear: the future of AI in education has to be human-centred, institution-led, and grounded in data governance that schools control.

The direction of travel is the same everywhere. Governments that are serious about AI in education are not buying products. They are building infrastructure.

The skills barrier is gone. That is the part that changes the game.

The reason schools have not done this before is obvious: they did not have the technical capacity. Building tools required developers. Developers cost money and schools do not have them.

That was true three years ago. It is no longer true.

The same AI capabilities that schools are trying to harness for students and teachers have eliminated the skills barrier to building with AI. A digitally confident Head of Department can now build a working feedback generator in a Google AI Studio session. A Deputy Head with some prompt engineering knowledge can create a bespoke report assistant tailored to their school's specific language and values. Tools that would have required a development team in 2022 can now be built by an educator with a clear problem and an API key.

And critically, the big players know this and are actively encouraging it. Anthropic launched free API credits for student-led educational projects alongside Claude for Education. They are not just selling to schools: they are seeding the next generation of school builders. The EdTech middleman is being quietly disintermediated by the very model providers the middlemen built their products on.

What forward-thinking schools are doing right now

The schools that will lead in 2030 are the ones making a specific decision in 2026. Not "which AI EdTech subscription should we buy?" but "what do we actually need AI to do in our school, and how do we own that?"

That decision looks like this in practice: securing a direct API agreement with a foundation model provider, understanding exactly what that means for data governance, identifying the two or three workflows where a bespoke AI tool would genuinely transform staff time or student experience, and building those tools around their specific curriculum, their assessment language, their school culture.

It is the difference between renting a generic tool and owning something built for you.

The schools doing this are not necessarily the most tech-savvy. They are the ones with leadership that has decided to stop inheriting other people's solutions to their problems.

This is where Floe Studio comes in

Darren Coxon and I built Floe Studio for exactly this moment.

We build bespoke AI tools for schools. You bring the API key, or we help you get one. We build the tool around your school: your curriculum, your assessment frameworks, your reporting language, your safeguarding requirements. You own it. You control the data. You are not dependent on a vendor's roadmap or a startup's survival.

The tools we build range from AI-assisted feedback generators and report writers to subject-specific tutoring assistants and staff productivity workflows. Every build starts with a conversation about what problem you actually need to solve, not what product we are trying to sell you.

If this resonates with where your school is heading, we would like to talk.

Get in touch at floestud.io