Contents

What's Inside

Chapter One · Facts checked June 2026

AI in the Classroom: Tools, Tutors & Analytics

Teaching is, I'd argue, the profession with the most to gain from AI — and the one with the least spare time to figure it out. So let's begin with a proper map. AI in education isn't a single technology; it's an ecosystem of specialised tools, and knowing the categories matters more than knowing any individual product.

The five territories

Educational AI clusters into five broad categories: adaptive learning systems (platforms like ALEKS and DreamBox that continuously assess and tailor content per learner), intelligent tutoring systems (AI simulating one-to-one instruction), content generation and assessment (materials, quizzes, feedback and grading support), learning analytics (dashboards and predictive models), and administrative tools (scheduling, communications, the paperwork that eats evenings).

The two-sigma echo

The intellectual lineage matters here. In 1984, Benjamin Bloom showed that students with one-to-one tutoring outperformed classroom peers by roughly two standard deviations — the famous "two-sigma problem", so called because individual tutoring at scale was economically impossible. Intelligent tutoring systems are the most serious attempt yet to close that gap: the research evidence shows effect sizes comparable to small-group instruction for well-designed systems in well-defined domains. Not two sigma, and not in every subject — but real, replicated and improving. Read the evidence cards in lesson 1.3 with your researcher's hat on; effect sizes vary enormously with implementation quality, which is rather the theme of this entire book.

Content generation: the first-draft engine

The most immediately useful category for working teachers: generative AI as a first-draft engine — lesson materials, differentiated versions of the same text, quiz questions, feedback comments, marking support for essays and code. The course's framing is the right one: AI content works best as a first draft for professional review, never as a finished product. Your subject knowledge and knowledge of your students is what turns a plausible worksheet into a good one.

Analytics: powerful, and handle with care

Learning analytics surface engagement patterns and flag students at risk — genuinely valuable for catching the quiet slide that busy classrooms miss. But every indicator in an at-risk model is a data point about a child, and the course's ethics warning deserves bold type: predictive flags are conversation starters for professionals, never labels for students. A risk score that follows a child around is not analytics; it's a reputation. We'll build the full guardrails in Section III.

Chapter Two · Facts checked June 2026

The Alpha School Case Study: Inside the 2-Hour Model

No school has generated more conversation about AI in education than Alpha School — co-founded by MacKenzie Price, expanding rapidly across the US, and built on a startling proposition: core academics, delivered by AI tutors, in two hours a day. This chapter examines it the way educators should examine everything: claims, evidence, trade-offs, lessons.

The model

The day splits in two. Mornings: a two-hour academic block of AI-guided mastery learning — each student on individualised tasks, progressing only on demonstrated mastery, with adults acting as motivators and coaches rather than lecturers. Afternoons: project workshops — life skills, entrepreneurship, public speaking, collaborative work — the things the model argues schools never had time for. The proprietary Alpha Learning System underneath tracks every interaction: responses, time-on-task and mastery state feed an engine that selects each student's next task in real time.

The claims, and the evidence test

The claimed outcomes are striking: top 1–2% test scores, a 3,000+ family waitlist in San Francisco, 90%+ student satisfaction. The educator's job is to hold two thoughts simultaneously: these results are interesting, and most rest on founder-reported data without independent verification — no published RCTs, selection effects everywhere (motivated families, premium fees, small cohorts). The course's evidence evaluator asks the only question that matters: what evidence would convince you? Independent replication with comparable populations is the honest bar, and it hasn't been met yet. Interesting ≠ proven, and a case study ≠ a blueprint.

The honest trade-offs

The model's tensions are real: gamification pressure (points and streaks motivating practice can shade into anxiety), limited unstructured time, heavy data collection on children, and the unresolved question of what's lost when direct instruction disappears. None of these are dismissals; they're the cost side of a ledger the marketing doesn't print.

What traditional schools can borrow

And yet — you needn't buy the whole model to learn from it. The transferable lessons: mastery-based progression (moving on when ready, not when the calendar says), data-informed coaching (teachers spending their scarce minutes where the dashboard shows they're needed), and the bracing question of time allocation — if AI genuinely compresses content delivery, what would your school do with the reclaimed hours? That question is worth a staff meeting all by itself.

Chapter Three

The Training Gap & the UNESCO Competency Framework

Here's the uncomfortable foundation under everything in this book: the technology has sprinted years ahead of the profession's preparation for it. UNESCO puts it plainly — "few countries have defined competencies or programmes to train teachers in AI." Students are using these tools daily; most teachers have had no structured training at all. That's the gap, and closing it is the most important AI investment any school can make.

The teacher–AI–student triangle

Why teachers first? Because every AI benefit reaches students through a teacher's choices: which tools, framed how, with what critical guardrails. An untrained teacher with a powerful tool is a risk; a well-trained teacher with a modest tool is an asset. Assess your school across the six readiness dimensions in the gap assessment — leadership, training, infrastructure, policy, culture, student preparation — and you'll usually find training is both the weakest dimension and the cheapest to fix.

The UNESCO framework: a shared map

UNESCO's AI Competency Framework for Teachers gives the profession its common language: 15 competencies across five dimensions — human-centred mindset, ethics of AI, AI foundations and applications, AI pedagogy, and AI for professional learning — each progressing through three levels: Acquire → Deepen → Create. Its quiet wisdom is what comes first: the mindset and ethics dimensions precede the technical ones, because why and whether matter before how. Use the self-assessment to map yourself across all fifteen; nobody scores "Create" everywhere, and the point is a personal development map, not a grade.

Literacy vs fluency

One more distinction sharpens all PD planning: AI literacy (conceptual understanding — what these systems are, how they fail, what they mean for learning) versus AI fluency (hands-on skill — prompting well, building materials, integrating tools into actual lessons). Both matter, and they develop differently: literacy through discussion and reading, fluency only through practice. The course adds a third idea worth pinning up in every staffroom: AI responsibility — using AI to extend thinking rather than offload it. That's the disposition we model for students, so it had better be the one we practise ourselves.

Chapter Four

Professional Development That Works

Knowing the gap exists is one thing; closing it without adding to anyone's workload guilt is another. This chapter is the practical PD playbook — phased, practical, and respectful of the fact that teachers are the busiest professionals alive.

The phased model: Awareness → Exploration → Application

SREB's three-phase model is the right shape for almost every school. Awareness: what AI is, what it means for education, the honest risks — delivered in twilight sessions and discussion, no hands-on pressure. Exploration: structured play — sandboxed time with approved tools, low stakes, shared discoveries. Application: AI woven into real planning, real marking support, real classroom practice, with peer observation and refinement. The fatal pattern the phases prevent: the single "inspirational" INSET day that leaps straight to application and leaves everyone overwhelmed by Tuesday. Plan the journey in the PD Phase Planner — terms, not days.

Pre-service and in-service

The same logic applies up and down the pipeline: teacher education programmes embedding AI competencies before qualification, and in-service PD meeting teachers where they actually are — which the readiness survey will tell you, if you ask. The course's eight ready-made workshop templates (filterable by duration and audience) save you designing from scratch; adapt rather than adopt.

Rubrics and confidence

Two quiet tools do disproportionate work. Lesson plan rubrics for AI-enhanced lessons make quality discussable — criteria for pedagogical purpose, critical use, data safety — so feedback is structured rather than vibes. And confidence-building is a deliberate design goal, not a hoped-for side effect: start with personal productivity wins (the email, the worksheet, the report comments) before classroom integration, because a teacher who has saved two hours believes in the tools in a way no presentation can achieve. Assess where you are with the readiness scorecard, and let people climb at their own pace.

Chapter Five · Facts checked June 2026

The Global Race: What Nations Are Doing

Wherever your school sits, it sits inside a national strategy — or a national vacuum. This chapter tours the world's approaches, because the patterns (and the cautionary tales) are remarkably instructive for decisions at every level, including yours.

Mandates vs guidelines

The global landscape splits along one axis: countries leading with mandates (compulsory curricula, national platforms, defined teacher standards) versus those leading with guidelines (frameworks, funding and ethics guidance, school-level discretion). The course's key insight names the tension honestly: the fastest movers gain scale and coherence but risk over-standardisation and privacy trade-offs; the careful movers protect trust but risk drift. Nobody has solved this; everybody is choosing a point on the line.

Around the world in five strategies

• China — the full mandate: the AI+Education action plan brings compulsory K-12 AI courses, a national teacher AI standard, government-backed platforms, and explicit 2030 goals. Unmatched scale; centralised everything.
• Estonia — the agile partner: AI Leap puts tools in front of 20,000 students and 3,000 teachers through direct partnerships with OpenAI and Anthropic, all inside EU GDPR guardrails. Small-country speed with big-firm capability.
• Singapore — the integrated platform: the Student Learning Space embeds AI modules nationally, governed by the MOE AIEd Framework's triple test of agency, fairness, safety — possibly the most exportable governance formula in this chapter.
• UK & US — market-led with public nudges: the UK's Oak Academy Aila assistant and targeted funding; the US's 2025 Executive Order, AI Education Task Force and national AI Challenge. Innovation-friendly, coherence-light.
• Finland — ethics first: national guidance built on transparency, bias-awareness and sustainable use, aligned with the EU Digital Compass and AI Act. The reminder that a coherent why can lead the what.

What travels

Strip the geography and the transferable patterns emerge: every serious strategy invests in teacher capability before student tooling; every durable one writes governance alongside deployment, not after; and the best pair ambition with honest evaluation. Sound familiar? It's this book's table of contents, at national scale. Your school can run the same play without waiting for the ministry.

Chapter Six

Human-Centred AI: Bias, the Digital Divide & Inclusion

Education's ethical bar is the highest of any field in this handbook series, for the simplest reason: the people affected are children, and participation isn't optional. This chapter assembles the equity framework — principles, bias, access, inclusion — that everything in Section IV must pass through.

The human-centred principles

UNESCO's anchor principle is that AI must "not widen technological divides" — and around it cluster the commitments shared by UNESCO, OECD and EU frameworks: transparency (students and parents can know how AI is used), accountability (a named human answers for AI-informed decisions), human agency (AI informs, educators decide), and inclusion by design. Run your school against the equity checklist; the gaps it finds are Section IV's to-do list.

Bias in the adaptive classroom

You met algorithmic bias in principle in Chapter One's warning; here's its education-specific anatomy. Training data bias: systems trained on data from some student populations serve others worse. Proxy variables: postcode, device type or writing style quietly standing in for demographics. Feedback loops — the most insidious in adaptive systems: a system that underestimates a student serves them easier material, generating performance data that confirms the underestimate. The mitigation playbook: demand vendor transparency about training populations, audit recommendations across student groups, and keep teacher override easy and culturally normal. A teacher's "this child can do more than the system thinks" must always outrank the algorithm.

The next digital divide

The divide is no longer just devices and connectivity — it's access to AI tools, and the skills and confidence to use them well. Schools that can't fund premium tools, families without home access, students whose only AI literacy comes from whatever the algorithm feeds them: this is the next gap, opening now. The infrastructure planner builds a phased bridge for your context — and the principle that guides it: when budgets force choices, fund equity of access before sophistication of tooling.

AI as an inclusion engine

And the genuinely hopeful flip side: used deliberately, AI is the most powerful inclusion technology schools have ever had. Real-time captioning and translation, text-to-speech and speech-to-text, reading-level adaptation, communication support for non-verbal learners — accommodations that once required specialist budgets now ship in mainstream tools. The discipline is making it deliberate: audit your toolkit's coverage across learner groups with the inclusion score calculator, and write equity safeguards into procurement — accessibility standards as contract requirements, not happy accidents.

Chapter Seven · Facts checked June 2026

Privacy, Copyright, Consent & Your School AI Policy

Now the legal spine. Schools process the data of minors at scale, which makes privacy law not a compliance chore but a duty of care with statutes attached. Four pieces: data protection, copyright, consent, and the policy that binds them into something staff can actually follow.

Student data protection

The frameworks differ — GDPR in Europe, FERPA in the US, national schemes elsewhere — but their demands on schools converge: lawful basis for processing, minimal collection, defined retention, and security proportionate to sensitivity. The operational rule that prevents most incidents fits on a sticky note: no identifiable student information goes into generative AI tools. Names, photos, assessment details, anything linkable — anonymise or abstain. Audit yourself with the compliance checklist, and use the framework matrix to know which rules bind you.

Copyright, both directions

Education's copyright question runs both ways. Outbound: student and teacher work is copyrighted, and schools must not feed it to AI systems for training without permission — that essay is the student's intellectual property, not training data. Inbound: AI-generated teaching materials sit in the same evolving legal territory as everywhere else — treat outputs as drafts you transform, keep human authorship meaningful, and follow the decision tree when cases get knotty.

Consent and transparency

When AI processes student data, someone must agree — and with minors, that means parents informed in plain language: which tools, what data, what purpose, what rights. Age-appropriate disclosure matters too; students deserve to know when AI is reading their work, in words they understand. The course's templates make this practical rather than noble: the consent flow builder generates jurisdiction-aware workflows, and the parent communication templates turn legal requirements into letters a human would actually read. Transparency done early builds the community trust that every later chapter spends.

The school AI policy

Everything converges in one document staff can follow on a busy Wednesday: scope, acceptable use (for staff and students), data handling rules, the vendor assessment requirement, incident response, and a review cycle. The policy generator drafts it section by section, and the weighted vendor scorecard handles the procurement half. The standing rule from every handbook in this series applies doubly in schools: a policy nobody reads is a liability dressed as a control — keep it to two pages, write it with staff, review it annually at most.

Chapter Eight · Facts checked June 2026

The Implementation Lifecycle: Pilots, Scaling & Infrastructure

Everything so far becomes real here: the phased journey from first pilot to embedded practice, with the gates, timelines and infrastructure that keep enthusiasm from outrunning readiness.

Four phases, with gates

The lifecycle runs pilot → evaluate → scale → integrate, and the discipline lives at the boundaries: a structured go/no-go gate at each phase transition, with criteria rated honestly before anything advances. The phase planner tracks the journey. The familiar warning from this series applies with school-shaped force: pilots without predefined success criteria become permanent fixtures nobody evaluates — define what "working" means before a single licence is bought.

Years 1–2: the pilot era

Short-term, keep it small and honest: identify priority needs (your Chapter Three gap report already did), select one or two platforms, and invest properly in teacher PD — the course's benchmark figures put pilots at roughly $5–10K in app licences plus ~$500 per teacher in training. Define the pilot in writing with the planning template: goal, target group, duration, success metrics, data collection, evaluation date. Modest and measured beats ambitious and vague, every term.

Years 3–5: scaling with evidence

Medium-term, the question changes from "does it work?" to "does it work everywhere?" The scale decision matrix rates pilots on weighted criteria before district-wide adoption; the integration checklist (27 items across six categories) tracks the deeper work of curriculum embedding. Two markers of seriousness at this stage: hiring or designating an educational technologist — the bridge role between pedagogy and platform — and budgeting for the unglamorous integration work that turns tools into practice.

Infrastructure: the floor everything stands on

None of it works without the basics: high-speed internet, modern devices, cloud services, and IT support that answers. The audit tool rates your school across six domains and generates a prioritised readiness report; the gap analysis template adds cost estimates. The equity note from Chapter Six echoes here deliberately: infrastructure gaps map onto disadvantage, so closing them is equity work, not just IT work.

Chapter Nine · Facts checked June 2026

Budgets, KPIs & Evaluation

Money and measurement — the two conversations that decide whether AI in your school survives its first leadership change. Both are easier than they look, provided you start them before anyone demands them.

The honest budget

The course's benchmark: a typical school AI implementation runs $20–50K in year one, depending on size, existing infrastructure and scope — with years two and three cheaper (licences and PD refresh) but never zero. Build it across all eight categories in the budget generator — and honour the ratio this series keeps finding: the training and people lines deserve more than institutions instinctively give them, because licences without capability are shelfware with a renewal date. Offset with the funding landscape: grants, government programmes and vendor education partnerships are plentiful right now; the vendor scoring tool keeps procurement honest while you shop.

KPIs worth tracking

Five categories cover the school dashboard: teacher competency (training completion, confidence scores), classroom usage (actual adoption, not licence counts), student outcomes (attainment, engagement), workload impact (hours genuinely saved — the metric staffrooms care about most), and equity indicators (access and outcomes across student groups — the category most dashboards forget and Chapter Six insists on). Build yours in the KPI dashboard builder, balanced-scorecard style, and set targets before the data starts arriving.

Evaluation: mixed methods, honestly applied

No single method tells the truth alone. Mix them: test scores and usage logs for the quantitative spine; surveys and classroom observations for the texture; controlled comparisons where scale allows; and ethical audits as a first-class method, not an afterthought. The evaluation plan builder assembles the design, timeline and data collection schedule. One cultural note worth enforcing from the start: evaluation exists to improve the programme, not to prosecute the teachers piloting it — say so explicitly and the data gets dramatically more honest.

Chapter Ten

Risk, Stakeholders & Your Action Plan

The final chapter closes the loop: the risks to manage, the community to bring along, and the one-page plan that turns this entire book into your school's next term.

The risk register

Education's AI risks are by now familiar friends — algorithmic bias, privacy breaches, widening inequality, student over-reliance, teacher resistance — and the management discipline is the standard one: rate each for likelihood and impact, map the heat, and attach mitigations with owners. The register template comes pre-loaded; the strategy picker sorts responses into quick wins, medium-term actions and long-term investments. The risk most schools underweight: over-reliance — students (and staff) offloading thinking rather than extending it. The mitigation is pedagogical, not technical: teach AI responsibility explicitly, as Chapter Three framed it.

Stakeholders: trust is the infrastructure

Map your community on the power/interest grid — leadership, staff, parents, students, governors, the local press if you're unlucky — and engage each quadrant deliberately with the map builder. The non-negotiables: transparency with parents (Chapter Seven's letters, sent before rumours fill the silence), training for teachers (Chapter Four, the engagement strategy that doubles as capability), and age-appropriate AI literacy for students — who are stakeholders, not subjects, and who will use these tools for the rest of their lives regardless of what school decides. Community trust is built in announcements and spent in incidents; bank plenty.

The one-page action plan

And so to paper. The action plan generator assembles everything this book has built — context, priorities, timeline, success metrics, stakeholder engagement — into a single page for your leadership team, and the 30-day starter checklist breaks the first month into daily and weekly actions so momentum starts Monday rather than "next term". You have the map of the landscape, the evidence habits, the global patterns, the ethical guardrails, the legal spine and the implementation playbook. What remains is the part no handbook can do: the first staff meeting where you put it all in motion. Teachers change the world for a living — this is just the newest set of tools for the job. Go well.