Preparing our families and children for a post-AGI world

Three independent lenses: what employers say (WEF), what adults can actually do (OECD), and which occupations AI is already used in (Anthropic, ILO, Stanford).

WEF 2025, fastest-growing skills: AI and big data, networks and cybersecurity, and technological literacy, alongside human skills such as creative thinking, resilience, flexibility and agility, analytical thinking and leadership. Employers expect the largest absolute growth in frontline, care and education jobs. 63% of employers cite skills gaps as a barrier to transformation, and 59 in 100 workers will need training by 2030 (WEF).

What the learning research says: design matters more than access

• Unguarded AI can hurt learning. In a field experiment with nearly 1,000 Turkish high-school students, unrestricted GPT-4 raised practice scores by 48% but cut later unassisted exam scores by 17%. A guardrailed "GPT Tutor" that gave hints, not answers, largely removed the harm (Bastani et al., PNAS 2025).
• Well-designed AI tutors can help a lot. In a Harvard physics RCT (N=194), students using a carefully designed AI tutor learned more than twice as much (median gains) as in an active-learning class, in less time (49 vs 60 minutes) (Kestin et al., Scientific Reports 2025).
• Low-cost tutoring at scale. A six-week after-school GPT-4 tutoring programme in Edo, Nigeria raised overall scores by 0.31 standard deviations (World Bank).
• Takeaway: ask kids to attempt the work first, and use AI as a Socratic tutor, not an answer machine.

What to emphasise at each stage. Where a line makes a factual claim it links to its source; the rest is practical judgement drawn from the evidence on this page.