A lot of my week has been one long reminder that AI is not just a feature you bolt onto a product. It's starting to change how entire systems work.
Not just schools, but health, justice, farming, sport, and the way we build software itself.
And when you zoom out like that, AI for schools stops being a classroom discussion and becomes a design discussion. What do we automate. What do we keep human. Where do we put care. And how do we keep people in the loop without drowning them in process.
Here are the threads I've been pulling on.
The 2035 future looks helpful and slightly weird at the same time
This Guardian piece paints a vivid, comic-style picture of what life in 2035 might feel like if AI becomes more capable across everyday services.
It describes AI taking a bigger role in routine healthcare, doing first-line triage and helping doctors weigh symptoms against medical history and huge volumes of research. It also talks about wearables and personal AI agents smoothing out daily routines with news summaries and small automated tasks, shaped by preferences and consent.
It then moves into justice, farming, and work.
- Legal prep and courtroom argument building could become heavily automated, speeding up cases and cutting costs.
- But it raises the obvious questions about bias and accountability when things go wrong.
- On farms, sensors and robots could enable earlier disease detection and better crop and soil decisions.
- In office work, productivity gains might make shorter working weeks more realistic.
- And it also flags something people don't talk about much: some people may struggle with more free time if work has been their main source of structure and purpose.
I don't read this as prediction. I read it as a prompt.
Because schools are where we teach people how to live inside whatever future arrives.
Read more: What will your life look like in 2035?
AI isn't a magic patch for old systems and that includes education systems
I was listening to Moonshots (guilty pleasure) and one comment stuck with me because I keep seeing it in real teams.
A lot of IT teams test AI the wrong way. They throw it at a legacy system and ask it to "fix it". Usually older stacks, lots of Java, sometimes even C. It struggles. Then everyone says "this isn't ready".
But the point isn't to patch the old thing. The point is to rethink it.
If you scrap the legacy approach, rebuild the workflow in a setup AI is strong at, and design the process for automation, it can move insanely fast.
Same with ops. Teams get stuck on small front-door problems like "it starts with an email in an Outlook folder and security gets in the way". Then they give up. But often you can fix that front end quickly, then let AI run through the rest of the process and crush the real work.
This is exactly how I think about AI for schools too.
If we keep asking AI to behave like a traditional tool inside traditional workflows, we'll keep being disappointed.
If we redesign the work around it, we'll be surprised by how quickly it can deliver.
Care beats content when you're trying to lift learning outcomes
The World Economic Forum video with Akshay Saxena (Avanti Fellows) was a solid reminder of something we often overcomplicate.
His takeaway from years working with underprivileged students is that learning improves most when students feel safe, supported, and part of a community.
Not just when they get better content or devices.
The video also makes a practical case for AI as support for teachers and caregivers, especially for:
- interpreting student data
- summarising case context
- strengthening counselling and motivation skills
The goal is not replacing relationships. It's reducing busywork so educators can spend more time on meaningful, human-to-human support.
This, to me, is one of the most grounded frames for AI for schools.
If the tool does not increase care, it's probably not worth the disruption.
Watch: Education and AI for the underprivileged
Governments are moving from "talking about AI" to funding it
Hong Kong is giving HK$500,000 to each public primary and secondary school that applies to introduce AI in classrooms, from a HK$500 million budget over three years.
Schools can spend it on AI tools and student AI literacy activities, but the plan comes with expectations:
- use AI in at least 3 subjects
- create teaching resources
- run demo classes and sharing sessions
- submit interim and final reports
I like that it's not just "here's money, good luck". It's tying the funding to capability building and sharing what works.
This is what scaling AI for schools looks like when it's taken seriously.
Read more: Hong Kong schools get HK$500,000 each under AI education plan
AI will change search and discovery, and education won't be immune
The AFR opinion piece on AI reshaping property search is about real estate, but the pattern applies everywhere.
If AI changes how people search, compare, and decide, it doesn't just disrupt a website. It pressures margins, business models, and expectations.
Key themes raised:
- AI tools directing people to options outside major portals
- higher tech investment needed to stay competitive
- the defensive strength of scale, data, and deep integration
- a shift from price increases toward cost control and smarter AI use
For schools, this matters because students and parents are already shifting how they "find answers". And if discovery changes, teaching needs to change too.
Not just "don't cheat", but "how do you judge quality, bias, and usefulness when search becomes a conversation".
Read more: Is AI coming for online property giants like REA?
Chips are destiny and it's worth understanding why
If you're interested in how AI chips are made, this CNBC video inside ASML's cleanrooms is worth a watch.
It's a rare look at the machines behind the world's most advanced chips and why lithography sits right at the centre of the AI supply chain.
A few bits I found useful:
- DUV lithography (193 nm) still does most of the work because it's reliable and scalable.
- EUV lithography (13.5 nm) prints the smallest features, but it's insanely complex (lasers, molten tin droplets, mirrors in a vacuum).
- High NA EUV is the next leap, enabling smaller features with fewer multi-patterning steps, improving yield on advanced layers.
- Leading edge chips aren't made with one technique. DUV, EUV, and High NA all work together across dozens of layers.
This matters for AI for schools because capabilities don't appear out of nowhere. They ride on infrastructure. Better chips means cheaper compute. Cheaper compute means more powerful tools become normal.
Watch: Inside ASML's cleanrooms - how AI chips are made
Real-time translation is quietly a game changer for learning and teamwork
Sydney FC trialling AI translation earbuds is one of those stories that feels small until you think about the second-order effects.
How it's being used:
- coaches speak once, players listen in their own language through an app
- one-on-one conversations without a human interpreter
- dozens of languages and accents
- players can speak in their native language, not limited English
Early impact sounds practical, not flashy. Clearer communication. Quieter meetings. Faster integration. More comfort expressing themselves.
There's still a delay and some domain language needs work, but it's already useful.
For AI for schools, the implications are obvious.
Multilingual classrooms. New arrivals. Parent-teacher conversations. Student support. Group work.
Reducing language friction changes who can participate fully.
Read more: AI earbuds shattering the language barrier in Sydney FC's dressing room
Agentic AI is pushing into knowledge work and that will flow into education
The OpenAI post about GPT-5.2 (aimed at professional knowledge work and agentic workflows) is another marker that we're entering a new phase.
The claim is basically: one system coordinating many tools across documents, code, data, and workflows, with stronger long-context reasoning and fewer errors, and different variants tuned for different use cases.
Whether you love or hate model release races, this direction matters for schools because:
- teacher workload is knowledge work
- leadership workload is knowledge work
- policy and compliance workload is knowledge work
- curriculum adaptation is knowledge work
If agentic systems become stable, they will show up in school operations and teacher workflows faster than people expect.
Read more: Introducing GPT-5.2
The thread I keep coming back to
Across all of this, one idea keeps repeating.
AI works when it respects the real world.
Not the perfect workflow. Not the ideal policy document. The real world.
- Schools run on relationships, not just content.
- Systems are messy, not clean.
- Workflows cross email, spreadsheets, phone calls, and "quick chats".
- Safety is about design, not just rules.
- Equity depends on whether we put care at the centre, not whether we buy shiny tools.
That's the test I keep using for AI for schools.
Does this tool reduce busywork and increase care. Does it make it easier for teachers to teach and students to learn. Does it help people participate more fully, including across language and access gaps. Does it keep humans in the loop in a way that's realistic.
If yes, it's worth leaning in.
And yes, I also made Christmas cards featuring my dogs
Anyone who's done a video call with me from home has met my crazy dogs Murdo and Niamh.
This year I made my own range of Christmas cards featuring them with Nano Banana Pro and they turned out super good.
Spoiler alert for my family.