Education in the Age of the Builder

The cost of building software is declining. Schools can now commission in a fortnight what took a procurement cycle a decade ago. Almost nothing in how schools buy, govern or evaluate technology has adjusted to that fact.

This series follows how technological progress cuts both ways. If school leaders can now build the tools they need, they also generate fragmentation faster as new tools fail to speak to existing ones. The deciding variable in the coming decade is therefore not capability but governance: who owns the data, who can audit the model, who holds the evidence, and who is able to say no.

To better understand the changes to education in this context, the essays draw upon the economist Mariana Mazzucato's conception of the common good. This framework replaces market-fixing with market-shaping where the terms on which public value is created are set deliberately in advance rather than regulated once that value has already left. A school can pass every procurement check in front of it and fail all five, which is a measure of how little procurement compliance tells a board.

Four chapters take one layer each: the learner with no single record, the educator whose workspace costs more time than it returns, governance with no unified intelligence or orchestration, and the edtech builders with no unifying integration layer. Each states the problem, sets out what the research shows, tests it against primary field research conducted with English multi-academy trusts, and specifies what optimal outcomes might be look like. 

The conclusion is that these are four layers of the same problem. Solving for individual stakeholders reproduces the fragmentation that caused the ever-present siloes that divides them.

The author is co-founder of Educave, an independent education research and technology company. The worked example in Chapter IV is Educave's own deployment and is flagged as such where it appears.

This is a conflict of interest rather than a disqualification, and the appropriate response is disclosure and method. The field research protocol and the basis for every figure are set out in the appendix. Where the argument would be commercially convenient, it is supported by an independent source or marked as untested. 

Nestled in the serenity of the Swiss mountains, Satya Nadella warned the Davos elite that "as a firm, if you are not able to embed the passive knowledge in a set of weights in a model that you control, by definition you have no sovereignty. That means you are leaking enterprise value to some model company somewhere." This argument should unsettle policy-makers and school leaders more than it did. 

Nadella's argument applies with greater force to institutions holding a public duty of care. Every essay submitted through a closed platform, every report drafted by a tool the school cannot audit, every subscription that ingests institutional knowledge and returns a dashboard is a small transfer of judgement out of the institution. Cumulatively, it is the upward transfer of an institution's greatest source of value: the ability to know itself and self-improve. 

The lack of sovereignty is symptomatic of a larger problem related to how the leakage became normalised and what should take its place. For that, it is useful to draw upon Karl Polanyi's observation that markets are instituted rather than natural [22]. A market is the outcome of a set of rules: what public bodies may buy, on what terms, and with what obligations attached to the purchase. Under this view, the fragmented education technology market is entirely intelligible according to the edtech procurement rulebook of the past few decades. 

The essays draw upon Mazzucato's framework in  The Common Good Economy [17]. The conventional framing treats public intervention as a correction: a market under-supplies something, the state patches the gap, then withdraws. Applied to education technology, that framing produces the settlement schools now have: picking winners, one-size-fits-all solutions, and light-touch evidence gathering. The common good perspective replaces market-fixing with market-shaping. Value in a public system is co-created by many actors, and the terms of co-creation are set deliberately in advance rather than regulated once the value has already left. 

Each principle is a testable property of an information system; whether:

- the system has a stated direction
- the people inside it shaped it
- the outputs can be audited
- the benefits reach pupils by need rather than by cost
- one institution learns is available to the next.

A school can fail all five while procuring impeccably, which is a measure of how little procurement compliance tells a board.

Why is this important?

Schools are not naive buyers. Rather, they are structurally disadvantaged. According to Hillman (2022), schools lack a full view of which vendors are credible, operate in a fast-moving and lightly regulated market, and frequently discover a tool was unfit for purpose only after years of use [1]. An increasingly structural condition of digital education is that procurement precedes evidence. If and when evidence arrives, the contract has renewed. This condition extends into every policy arena. 

Indeed, my introduction to this self-reinforcing (or self-justifying) issue was in the think-tank industry where taxpayer-funded international development projects were renewed without methodologically rigorous monitoring and evaluation mechanisms. Research is a tool to game the procurement system to secure contracts which benefitted institutions and technocrats in the Global North more than the supposed benefactors of the education development programmes in poor countries. In this policy arena, these outcomes are perpetuated by financial and corporate institutions incentivised to prolong military sales and operations. It comes as no surprise to the A Level Politics students who I've mentored when we evaluate institutional knowledge production that think-tanks and lobby groups are enmeshed with defence contractors in what Eisenhower called the military-industrial complex. In such profitable markets, business is business. 

Similarly, in Edtech, Hooper, Livingstone and Pothong (2022) found that Google Classroom and ClassDojo operate under terms in the UK which mean schools carry data protection responsibilities they are not resourced to discharge, cannot meaningfully negotiate standard terms, and lack the technical visibility to know what is collected or where it goes [2]. Crucially, Hooper et al argue that these companies or products aren't uniquely bad. Rather, the governance framework around them assumes an institutional capability that does not exist. In common good terms, this is a transparency failure with a specific location: the controller cannot see what it is accountable for.

The principle of Co-Creation and Participation fails in the same way. Ryan-Atkin's longitudinal study of four multi-academy trusts in northern England found local governors moved from policy actors to policy subjects; one described the role as very much a rubber stamp[18]. Local authority representatives had withdrawn from the boards that remained, producing both a democratic deficit and a knowledge deficit about the communities served [18]. Allen and Gann argue that legitimacy in an academised system has to be rebuilt through co-creation because the architecture no longer unifies stakeholders [19]. In terms of the impact on digital infrastructure, a trust that has centralised decision rights without centralising visibility has removed the local check and not replaced it with anything a board can evaluate its strategic objectives against and decide what to do next in the most timely way possible.

UNICEF's Innocenti office reached a comparable position in its 2025 review of data governance in education technology. The authors argue that children's education data requires governance designed around the child's interest rather than the vendor's convenience, and that clear, robust rules serve innovators too by defining the boundary they can build inside [4]. The Digital Futures for Children research programme finds that transparency from data controllers coupled with agency for data subjects is the most plausible basis for a public-service data ecosystem [3].

Data sovereignty is increasingly important in the context of AI-assisted cyberattacks. Education is now attacked more consistently than almost any other part of the public sphere; largely through systems it did not build, does not own, and cannot fully audit. The Department for Science, Innovation and Technology's Cyber Security Breaches Survey for 2025/2026 found that 98% of higher education institutions, 88% of further education colleges and 73% of secondary schools identified breaches or attacks in the preceding twelve months, against 43% of businesses; 29% of universities reported incidents at least weekly [12]. 

None of this is new. In the mid-2000s IMS Global, now 1EdTech, published the Learning Tools Interoperability specification: a common protocol so that tools from different vendors could exchange data and identity [13]. The rationale was related to public-interest. Open standards make systems more valuable to the institutions that run them, not just the firms that sell them.

LTI succeeded narrowly and failed broadly. While it is among the most widely adopted standards in the sector, fragmentation is worse than when it was written. The OECD's assessment two decades later is: 

- Many digital education tools still cannot exchange or link data
- Staff re-enter the same information repeatedly
- A significant share of system-level data remains dormant rather than being used for improvement or intervention [6].

Most countries have not achieved interoperability between school-level and system-level management tools, and in some cases not even between system-level tools [6]. This suggests that adoption does not create the incentive to share. Therefore, a standard is a suboptimal governance regime

Nor does the arrival of a more capable technology dissolve the problem. The OECD's 2026 Outlook examines generative AI across four distinct settings: students learning subject knowledge, students and teachers together in instruction, teachers working alone, and institutional and system functions such as the analysis of learning pathways and study advice. It treats the opportunities and the concerns as inseparable [20]. That separation of use cases is the useful contribution. It makes clear that there is no single question called does AI work in education, and therefore no single procurement decision that answers it. Direction has to be set per use, per institution, by someone accountable for the answer.

Mazzucato et al apply the same logic to artificial intelligence, and their central distinction is between the rate of innovation and its direction [21]. Policy attention concentrates almost entirely on rate: adoption, competitiveness, capability. Direction is left to whoever builds fastest, which has often meant the dominant market players who may be looking over their shoulder at increasingly agile, AI-assisted product teams. The authors also name the mechanism by which public bodies lose the capacity to govern what they buy. Sustained outsourcing hollows out internal technical judgement until the institution can no longer specify, evaluate or challenge what a vendor supplies. Read against the field research in this study, that description is exact. The trusts we spoke to are not short of platforms. They are short of the internal capability to interrogate them, and that capability is the thing procurement rules do not ask them to hold.

This produces the counter-intuitive risk of the Builder Age. If the cost of building falls and schools respond by building more tools without a shared governance layer, fragmentation does not improve. It compounds: faster, cheaper per tool, with more data in more places and more surface to attack. The age of the builder without interoperability and good governance accelerates education systems towards suboptimal outcomes for teaching, learning, and governance teams. 

The four chapters that follow are the four places this becomes concrete: a learner with no single record, an educator whose workspace costs more time than it returns, a governing layer with no orchestration, and an evidence base thin enough that a sector spends billions on faith. They are one problem. Solving any of them separately reproduces the fragmentation that caused them.