This week's reads felt like a reminder that "AI in education" is no longer one conversation.
It's becoming a stack.
At the top you've got national plans, compute, exports, and public trust. In the middle you've got foundations models, agents, and platforms that make AI usable inside real workflows. And at the ground level you've got the classroom reality. Teachers needing support. Students needing real AI literacy. And schools needing tools that fit education, not generic products awkwardly repurposed.
Here are the pieces that shaped my thinking this week.
Earth AI shows where multimodal AI is heading
Google has introduced Earth AI, a family of geospatial AI models and reasoning agents that combine imagery, population, and environmental data to provide insights at planetary scale.
What stood out is the "cross-domain" nature of it. This isn't just analysing a satellite image. It's combining signals to answer messy real-world questions, like identifying communities most at risk from storms or floods, and supporting faster disaster response.
Key updates mentioned:
- New foundation models for satellite imagery and population dynamics
- A geospatial reasoning agent powered by Gemini models that can plan and execute analyses across datasets
It's a strong example of what happens when multimodal models meet a real problem space with real data.
Read more: Google Earth AI - unlocking geospatial insights
Australia is now talking openly about where we could specialise
Australia's AI Opportunity Report outlines how AI could lift growth, wages, and exports by 2030.
The headline numbers are big:
- Up to $142bn per year in total economic value by 2030
- Up to $112bn per year from broad adoption alone
- A further ~$18bn per year from domestic capabilities
- And ~$11bn from exports (including applications, compute, and education)
It also calls out a point I think is under-discussed. If we focus only on adoption, we leave a lot on the table. The value chain matters. Compute, applications, workforce, research translation, and trust all matter.
A few "what it will take" items in the report that feel very real:
- Clear national choices on where to specialise
- Faster approvals and investment for data centres and networks
- Upskilling, with managers and professionals expected to spend significant hours using AI tools
- Reversing the decline in AI research and translating it into capability
- Building public trust in safe, responsible, democratic AI
Report link: Australia's AI Opportunity Report
Beijing making AI education compulsory is a signal to the world
Beijing has become the first region in China to make AI education compulsory across all public primary and secondary schools.
More than 1,400 schools are offering structured AI lessons each semester, giving students hands-on experience with coding, robotics, and large language models.
Whether you agree with every detail or not, it's a clear signal. Some systems are moving from "AI as a tool" to "AI as a literacy".
I also caught up with Claire Butler this week and loved her framing. The challenge isn't whether students can still think critically with AI. It's whether we design learning that deepens reasoning while building authentic AI literacy.
Used well, AI can create a kind of cognitive lift. Helping students interpret evidence, weigh ideas, and reflect on their own thinking.
Read more: Beijing makes AI education compulsory in public schools
Teachers won't be replaced but teaching will change
In Education Week, Ingrid Guerra-López makes a point I strongly agree with.
The real question isn't whether AI will take over classrooms. It's how educators use it to enhance their work.
Used well, AI can support more one-to-one feedback and better differentiation for diverse learners. But none of that happens without investment in teacher preparation and ongoing support.
New tools change pedagogy, not the purpose of teaching. The goal stays the same. Helping students think critically, create meaning, and grow.
Read more: AI won't replace teachers, but teachers who use AI will change teaching
A new study shows why "five-word prompts" don't produce great lesson plans
I've said before that expecting AI to produce high-quality lesson plans from a five-word prompt is unrealistic.
AI needs context.
A study analysed over 300 AI-generated lesson plans from ChatGPT, Gemini, and Copilot and found:
- 90% of activities focused on lower-order thinking (memorising or summarising)
- Only 6% included multicultural perspectives
- Few lessons promoted analysis, creativity, or civic engagement
The authors recommend what most good teachers already know: AI should support teachers, not replace them. And better outputs require better inputs, including student needs, learning objectives, and the frameworks that matter in a real classroom.
Read more: AI-generated lesson plans fall short on inspiring students
Why AI built for schools is different to generic AI
Some people have asked why AI made just for schools is even needed.
Here's the simplest version.
When AI is built for education, it can carry context by default:
- no need to upload extra info or write a long prompt
- it already understands the school setting, so the first answer is better
- it can align to curriculum, progression, and classroom realities without you fighting it
That's what happens when AI is built for education, not just adapted to it.
I'm currently testing the next version of CurricuLLM that takes context awareness to a whole new level. I'm hoping to reveal it in early December, but I'm so excited I might share early glimpses with anyone who reaches out.
(Also, is there a seahorse emoji?)
Miro's update is a good example of "AI that fits the workflow"
Miro announced a set of new AI features recently and I think it's one of the more notable releases, mostly because it fits the product people already use.
The tools make it easier to summarise, cluster, and build on content inside a board without forcing a new workflow.
Highlights include:
- AI Assistant for summarising and organising board content
- Auto-clustering of sticky notes and ideas
- Smart prompts to generate templates and actions
- Early tools for lightweight apps and automations
Out of all the low-code, agent-ish platforms flying around lately, this feels like one of the most likely to catch on with citizen developers. It blends naturally with collaboration instead of demanding a whole new system.
Visit Miro Innovation Workspace Watch: Miro AI features demo
What I'm taking from this week
The direction is pretty clear.
AI is moving from "ask a chatbot" to "systems that understand context, plan work, and operate inside real environments".
For AI for schools, that shift comes with a responsibility.
- Build AI literacy that includes critical thinking and social understanding, not just "what is a model"
- Give teachers practical support and time, not just tools
- Design for higher-order thinking, not default summarising
- Prefer systems that carry context safely, so teachers don't have to fight prompts all day
- Keep the purpose of schooling intact, even as pedagogy evolves
The tech is accelerating.
The real work is making sure learning gets deeper, not just faster.