Education & Families in the Age of Agency

Executive summary

• AI in learning is now a programme of record, not a pilot. Governments in Europe, Asia and the UK are funding AI tutors and assistants at national scale ^[3][4][5]. At the same time, AI agents can now complete whole online courses for students ^[7]. Leaders face this tension now, not in some later phase.
• The evidence favours how AI is designed, not how much of it there is. The OECD finds that students using general-purpose chatbots often turn in better work, but the gains disappear or reverse once the AI is removed. Tools built around teaching goals show more lasting benefit ^[1][2].
• Regulation now centres on children and is already in force. Australia's under-16 social media rules, the UK's Protection of Children Codes, California's companion-chatbot law and a US FTC inquiry all put duties on AI services that young people use ^{[11][13][14][15][16]}. The EU's high-risk rules for education AI arrive in December 2027 ^[17].

What's changing

National AI tutors are going live. Estonia's AI Leap gave 20,000 upper-secondary students and 3,000 teachers access to AI chatbots from September 2025. It plans to add vocational schools and a further 38,000 students from September 2026 ^[3]. In January 2026 England's Department for Education said AI tutoring tools could reach up to 450,000 disadvantaged pupils in Years 9–11. Teachers start co-designing the tools in summer 2026, trials follow later in the year, and schools are to have access by the end of 2027 ^[4]. Singapore's Student Learning Space includes an AI Learning Assistant that gives students real-time feedback on their writing. It is designed to redirect students who ask for direct answers instead of supplying them ^[5].

Platforms are adding agents to everyday school tools. At Bett 2026 Google announced Workspace Studio, a tool for building and sharing AI agents that automate work, with new ways for education users to access it. It also expanded Gemini across Classroom and Chromebooks ^[8]. Systems are moving from answering questions to carrying out tasks inside the tools schools already use.

Students' agents are acting too, sometimes against the institution. The New York Times reported in 2026 that AI agents are logging into learning platforms to watch lectures, sit quizzes and post in class discussions on a student's behalf. The major learning-management vendors say no block is foolproof ^[7]. Coursework done unsupervised can no longer be assumed to show what a student can do.

Agentic care is entering the home, but cautiously. Australia is assessing AI in aged care, including robot companions, behaviour monitors and pain-management apps. Current uses are mostly documentation scribes, translation support and help with clinical assessments. A government pilot will involve only up to 20 professionals ^[9]. In the US, providers of services for older people have asked federal regulators for clear oversight, required human review and the right to appeal AI-influenced decisions ^[10].

Some reversals are a warning. In August 2025 South Korea's National Assembly removed official textbook status from AI digital textbooks. Adoption had stalled at around 30 per cent after a rushed rollout, leaving publishers and schools with sunk costs ^[6].

Through the Age of Agency lens

Age of Agency: decide where AI may act, and who can override it. Once a tutor sets the next task, or a care monitor raises an alert, the institution has delegated a decision. Boards need a delegation map that states which actions agents may take on their own, which need sign-off from a teacher or carer, and which are off-limits. Singapore's age-staged model is a workable template. Pupils in Primary 1–3 do no direct AI work. Pupils in Primary 4–6 use only education-specific tools under teacher supervision. Secondary students may use AI where permitted, and must declare it ^[5]. A school that governs delegation explicitly will make better decisions than one that leaves it to whatever its vendors set as defaults.

Agentrification: value shifts from producing work to judging and verifying it. As agents take over drafting, marking support and administration, educators' work moves towards designing learning, authenticating it and supporting students' wellbeing. The OECD's TALIS data show 37 per cent of lower-secondary teachers already use AI, and 72 per cent worry about academic integrity ^[2]. Assessment is the role under most strain. When agents can complete courses ^[7], credentials lose value unless institutions invest in supervised, in-person and oral checks. The winners will be providers whose credentials still prove what a student can do. Online-only programmes that rely on unsupervised assessment carry the greatest exposure. In care, the same shift moves workers' time from paperwork to people, provided the scribes and monitors are clinically validated ^[9][10].

Universal Basic Intelligence (UBInt): treat capable AI tutoring as public infrastructure. England's choice to aim its tutoring programme at pupils on free school meals ^[4], and Estonia's universal provision ^[3], both treat AI tutors as a public good rather than a premium add-on. This matters because UNESCO warns that AI without safeguards can deepen inequality. It notes that around 2.6 billion people lacked internet access in 2024, and that student data is being commercialised ^[18]. Under UBInt, the question for schools and systems is who receives a capable, safe, teaching-grade agent, and on what terms, not whether to adopt AI at all. A free general chatbot is not the same thing. The OECD evidence suggests purpose-built tools, not general assistants, deliver learning that lasts ^[1][2].

Risks and governance

Child safety and companionship. In a 2025 Common Sense Media survey, 72 per cent of US teens had used AI companions, and a third of those users said a bot had made them uncomfortable ^[19]. In the UK, 71 per cent of children say they have used an AI chatbot. Researchers say current regulation does not adequately cover over-reliance or emotional dependency ^[14]. Australia's eSafety Commissioner has warned about social engineering risks when companion technologies take on caring roles for older people ^[9].

Over-reliance. The OECD warns that handing thinking over to chatbots can cause "metacognitive laziness", which erodes long-term skill building ^[2].

Privacy and data. UNESCO flags that private platforms use student data for advertising and to train AI models, and recommends privacy-first practice ^[18].

Integrity. Agents posing as students ^[7] put assessment validity, and with it accreditation, at risk.

The regulatory picture:

• Australia: Since 10 December 2025, platforms must take reasonable steps to stop under-16s holding social media accounts, with fines of up to A$50 million ^[11][12]. From 9 March 2026, an industry code requires age assurance, parental controls and safe defaults for social media services, including those with AI companion features ^[13].
• United Kingdom: Ofcom's Protection of Children Codes under the Online Safety Act have applied since 25 July 2025 ^[14].
• United States: In September 2025 the FTC issued orders to seven companion-chatbot operators, seeking information on safety testing, age controls and engagement design for minors ^[15]. California's SB 243, in force since 1 January 2026, requires AI disclosure and break reminders for known minors, and allows people harmed by violations to sue ^[16]. Executive Order 14277 points federal grants towards AI literacy, teacher training and AI-supported tutoring ^[20].
• European Union: The AI Omnibus moved the start of high-risk obligations for stand-alone Annex III systems, which include education, to 2 December 2027 ^[17]. Procurement should assume those obligations apply now.

Questions for boards and leadership

1. Which decisions about learners or care recipients do we let AI agents make or start today, and who can override them?
2. Do our AI tools hold back answers to build skill, or do they simply produce work? Do we know which kind we have bought?
3. If an agent could complete our unsupervised assessments, which of our credentials would still prove what a student can do?
4. Do our vendor contracts set out age assurance, data use, model training on student data, and exit terms? Would they meet the EU's 2027 high-risk standard?
5. Who in our organisation is accountable for children's emotional safety when they use AI, including tools they bring from home?
6. Are our most disadvantaged learners and families getting the best AI provision, or the cheapest?
7. In home and aged care, which AI uses have been clinically validated, and where is a human required to review?

How KIG helps

• KIG Advisory. Strategy and transformation for schools, universities, edtech companies and care providers. This covers delegation maps for agents, assessment redesign, governance for vendors and procurement, and readiness for regulation across Asia-Pacific, the US and Europe.
• KIG Research. Applied frameworks covering the Age of Agency, Agentrification and UBInt, plus published research on adolescent digital wellbeing, turned into decision tools that boards and school leaders can use.
• KIG Ventures. What we learn from building and backing AI companies in digital wellbeing, early learning and senior care, applied to practical questions of product safety, evidence and how to scale responsibly.