A few different threads this week, but they keep circling the same question: how do we bring AI into education in a way that actually holds up? That runs from how we teach AI literacy, to how communities push back, to what the market is really selling, to what a system does when it stops pretending the tools are not already in students' hands.
Three models for teaching AI literacy
A recent piece sets out the three models Canadian provinces are using to teach AI literacy in K-12, and the trade-offs map neatly onto the choices any system faces. A dedicated subject gives you protected time, clearer sequencing and real assessment, but it often reaches only the students who opt in. Embedding AI in existing subjects ties it to real problems, but the time and training rarely follow the new content. A transversal framework woven across every subject reaches every student, yet without clear progression markers it depends on local enthusiasm to land.
None of these is the obvious winner, and I suspect most systems will end up with a combination: anchored in curriculum policy, built with teachers rather than at them, and backed by real investment rather than goodwill. Australia is heading into exactly this conversation, and it is worth being honest that each model asks for something we do not always fund.
When withdrawing a school is the system working
New York recently withdrew plans for an AI-focused high school after parent pushback. The proposal moved fast, families said the process felt rushed and opaque, and officials listened and pulled it.
It is tempting to read that as a failure, but I think it is closer to the opposite. Technology changes fast; people and communities move at human speed. That gap tends to get framed as friction, yet it is doing real work. It is where parents ask hard questions about cognition and mental health, where admissions policies get scrutinised, and where "emerging technology" stops being a slogan and becomes a decision about actual children. The slow part is not an obstacle to getting this right. It is how we get this right.
What education really sells
Coursera's Q1 2026 numbers look strong at a glance: revenue up, AI course enrolments running at 20 per minute. Look closer and the interesting detail is the mix. Professional Certificates tied to specific employers grew enrolments 91 per cent, while full degrees, the most gated product, are now an explicit drag on growth.
Set that against the wider field. Udemy posted its first-ever revenue decline, 2U/edX went through Chapter 11, and Chegg has lost 99 per cent of its market value to ChatGPT. The pattern is hard to miss. The product was never the content. The product is verification, and verification is really just trust. AI can generate infinite content, but it cannot yet be trusted to vouch for anyone. That gap is what the survivors are quietly selling, and what the collapsing players forgot they were selling in the first place.
Estonia's national AI pilot
A 2024 Estonian national survey of more than 15,500 people found that up to 90 per cent of students were already using freely available AI tools for schoolwork, compared with 53 per cent of teachers. That gap alone tells you where the pressure sits.
Estonia's response was to launch a EUR 4 million pilot across 154 schools, involving 20,000 students and 4,900 teachers, supported by regular teacher training and school-based learning communities. What I find compelling is not the budget but the posture. It is one of the clearest examples of a system responding to reality rather than pretending the tools are not already in students' hands. The training and the learning communities matter as much as the money, because they are how you close the gap between what students do and what teachers feel equipped to guide.
Non-coders shipping real products
For years, educational innovation ran on a familiar loop: a teacher had a brilliant idea, everyone agreed it was fantastic, and then everyone waited years for a vendor to build it. Generative AI is collapsing that loop.
I keep coming back to accounts from educators building real tools themselves, assembling what looks suspiciously like an AI engineering team and discovering that "vibe coding" is both genuinely powerful and mildly terrifying. It captures where this is heading. AI here is far less about replacing the people who know what good looks like than about amplifying them. The judgement, the sense of what students actually need, still has to come from someone who has stood in a classroom. Nothing excites me more about the next five years than giving creative people the power to bring their ideas to life.