Experimental futuring: AI prototyping and pilot testing as anticipatory governance in education

Synthesis: Williamson and colleagues analyze how England's Department for Education (DfE) turned future-oriented talk about AI into policy machinery between 2023 and 2026, naming the resulting mode of anticipatory governance "experimental futuring." The study is documentary analysis of policy texts, not an experiment or trial of its own: the authors read DfE and partner documents as "tools of governing" that make governance ambitions actionable. They argue that proof-of-concept prototyping and live school pilot testing are used to prefigure a particular AI future for schooling, sediment promissory claims about edtech scalability, codify the curriculum for machine readability, and reconfigure teaching — while foreclosing democratic or participatory alternatives. Read carefully, this is a critique of how such AI Governance is designed to make one AI future appear inevitable, not evidence that it works.

Key Findings

1. The paper conceptualizes "experimental futuring" as a form of anticipatory governance: designing and engineering AI prototypes and pilot testing programs intended to model, prefigure and govern the pedagogic possibilities of AI in the future.
2. The method is in-depth documentary analysis of DfE and partner policy materials from spring 2023 to summer 2026, read through a "critical futuring" lens; the authors treat the texts as performative instruments that attract interest and animate actions toward a defined future.
3. Prototyping is analyzed as pedagogic prefiguration. The DfE and DSIT announced a £4 million fund for an educational "AI Content Store," with £3 million awarded to the contractor FacultyAI and £1 million allocated to 16 edtech companies through an AI Innovation Fund to build working prototypes.
4. The program extended in 2026 into an AI Tutoring Pioneers Program led by the Incubator for AI (i.AI) at DSIT, committing up to £300,000 per organization for each of eight competitively tendered developers of prototype AI tutoring applications.
5. Pilot testing is analyzed as promissory evidence production. The DfE's EdTech Impact Testbed Pilot committed more than £1 million in 2025; in January 2026 the Education Secretary announced an additional £23 million to expand it into a four-year program recruiting 1,000 school sites, with the Education Endowment Foundation named "evidence partner."
6. The authors argue this shifts evaluation from a scientific synthesis of "what works" toward just-in-time evidence about what "promises to work," adopting the government's "Scan > Pilot > Scale" process model and treating classrooms as live testbeds that are modified to become more observable, analyzable and influenceable.
7. The program is read as prototypical of a wider reorganisation of the English public sector for AI testing, with the education Content Store foreshadowing a "National Data Library" and England positioned as a scalable reference model for other jurisdictions.

Prototyping as pedagogic prefiguration

From spring 2023 the DfE worked with public and private partners to shape AI in schools. FacultyAI, a favored government contractor, ran a teacher hackathon in 2023 alongside the National Institute of Teaching, and claims about AI freeing teacher time and personalizing content were framed as evidence justifying further experimentation. In summer 2024 the DfE and DSIT announced the £4 million Content Store fund — a repository of curriculum guidance, lesson plans and anonymized pupil work meant primarily as a reference for companies to train AI models. Its three prototypical components were a store of structured, tagged educational data, a prototype explorer app, and a prototype API; it later entered "public beta."

The authors treat a prototype not as a neutral technical object but as a relational composite of future imaginaries, actors, designs and institutions that "respecifies" work once it enters a social site. On this reading the Content Store does not merely support teaching; it reshapes it. A proof-of-concept technical report specified the need for "codification of the National Curriculum" into "AI-friendly structures" — tagging learning objectives, error types and correction suggestions in standardized formats machine tools can process — so that to work as intended, curriculum materials must first be made machine-readable. The teacher is thereby repositioned as a secondary reader and user of materials designed for machine interpretation, an "augmentable" professional whose autonomy is delimited by the technology's capacities. DfE product safety standards issued in early 2026 function further as an accountability mechanism keeping prototype applications within the Department's mandated bounds.

Pilot testing as promissory evidence production

If prototyping prefigures practice, pilot testing is where the DfE seeks to generate evidence for scaling. The 2025 EdTech Impact Testbed Pilot "aims to identify and evaluate innovative educational technologies that can enhance teaching and learning and reduce workload in schools and colleges across England," and sought to build the evidence base on the impact and scalability of promising technologies. A testbed is defined by the DfE as "a controlled environment used for testing and evaluating technologies or software applications," with three aims: identify edtech tools, generate evidence, and test the testbed itself. A methodology report followed late in 2025.

The program then scaled rapidly. At an education technology industry event in January 2026 the Education Secretary announced a further £23 million to expand the pilot into a four-year program recruiting 1,000 school sites, followed by a call for AI Tutoring Pioneers to pitch technologies for testing there. The Education Endowment Foundation was named evidence partner to "build a stronger evidence pipeline," starting with rapid evaluations and later large-scale evaluations of the highest-promise programs. The authors stress that this pipeline produces promising signals from early-stage prototypes rather than settled findings, and that the Tony Blair Institute's reports ("Generation Ready," September 2025; "Scaling Safe, High-Quality AI in England's Schools," January 2026) push an even more explicit agenda of a permanent national testbed network, a national AI-testing lab, and mass "AI diffusion."

Schools reconfigured as testbeds

Drawing on a sociology of testing, the authors argue that classrooms have become sites of "testing in the wild," where technologies are introduced early and user feedback reveals technical and ethical problems. Testbeds do not simply test within a setting; they modify the setting. Schools become "living laboratories," made more observable, analyzable and influenceable, with teachers and students refigured as test subjects and the software engineer, consultant and policy maker substituting for the scientist's expertise. The testbed "makes human life itself an experiment for technology futures," and what results from the test is a reinscribed expectation of the future classroom — an automatable, AI-augmented space in which teachers take subsidiary roles.

A mutation of the evidence-based model

The paper's central conceptual claim is that experimental futuring amalgamates three strands: new collectives of state and private actors experimenting with futures; engineering-based prototyping as future-making; and the international evidence-based movement's experimental methods of evaluation. By invoking "evidence based tech," the DfE connects its program to the apparent rigour, objectivity and authority of that movement. But rather than mobilising a settled evidence base or longitudinal trials, the authors argue, the Department has moved toward speculative, pre-emptive policy informed by just-in-time promissory evidence from live prototype tests — a mutation that challenges the nature of evidence used in public policymaking about AI, and that resembles the technology industry's "tech business experimentalism."

What this means for practice

• School leaders and administrators. Ask who owns the evidence a product pilot produces, whether participation gives you any leverage over how results are used, and whether joining a testbed fits your school's priorities or mainly supplies demonstrations for national scaling.
• Teachers. Treat pilot participation as a change to working conditions as well as a tool trial: prototypes that require standardized, machine-readable materials respecify lesson planning and assessment, so negotiate what is expected to become routine.
• Policymakers. Distinguish promissory evidence from validated effects. A "rapid" pipeline that reports promising signals is not a substitute for robust evaluation, and the paper argues it can secure consent for a single pathway while foreclosing alternatives.
• EdTech developers and researchers. Note the ethical gap the authors identify: live testing without oversight of functionality may affect students' and teachers' rights, and trials can render innovations unavailable for independent critical evaluation.
• Institutions. Build in deliberation — participatory horizon scanning, future design, citizens' assemblies — so that choices about AI in schools are debated rather than settled by engineering demonstration.

Limitations

• The study analyzes the program architecture and policy documents only; the authors explicitly state it does not engage with on-the-ground enactment in schools, and that impacts on actual school practice would require further empirical analysis.
• It is a single national case (England), so its findings are context-specific even as the program is positioned as a scalable reference model internationally.
• The authors acknowledge that the DfE program is "no doubt messier in practice than it is presented," and that their focus on documents, not implementation, brackets that messiness.
• The paper is a critical, conceptual analysis rather than an empirical test; it does not measure effects of AI use in classrooms and cannot speak to learning outcomes.
• The privilege of participating schools — those with the institutional leadership, resources and networks to act as implementers — is inferred from the program design rather than examined through data on which schools took part.

Citation

Williamson, B., Facer, K., Pykett, J., Sriprakash, A., & Valladares-Celis, C. (2026). Experimental futuring: AI prototyping and pilot testing as anticipatory governance in education. Globalisation, Societies and Education.