A bunch of threads came together for me this week.
Some are flashy, like AR glasses and "agentic browsers". Some are boring-in-a-good-way, like procurement guidance and classroom guardrails. And some are just reality checks, like how much cleanup work sits behind "vibe coding".
Put together, it feels like we're moving from "AI as a tool you open" to "AI as a layer that sits around you", in your devices, your browser, your workflows, and eventually your classrooms.
Glasses that actually do something useful
Chunky glasses may not be a fashion statement, but they might not need to be.
Meta has announced the first Ray-Bans with a built-in augmented reality display. The idea is simple. Put useful information inside your lens, not on your phone.
The feature list is basically a preview of "ambient AI":
- real-time translation and captions
- turn-by-turn directions
- camera viewfinder, photo and video sharing
- an AI assistant for questions in the moment
- six hours of use, plus a charging case
They also integrate voice, touch, and a Neural Band wrist device for hand gestures.
If this lands well, it changes the default "shape" of computing. Less screen time. More context. More assistance in the flow of life.
Link: Read more: Meta Ray-Ban smart glasses with AR display
Agentic browsers will turn the internet into a coworker
You're going to hear "agentic browser" a lot more next year.
A normal browser shows you pages and waits for clicks. An agentic browser listens to what you want to achieve and does the steps for you.
Instead of working through forms field by field, you describe the outcome and the agent completes the process. It can move across systems, pull data, join results, and prepare an output you review.
If that sounds like RPA, that's because it is the next step of the same idea. The big difference is that the agent can handle messier interfaces and language, and can adapt as it goes.
The promise is obvious. Less busywork. Faster end-to-end processes.
The risks are also obvious:
- agents can make mistakes
- sites block automation
- prompt injection and data exposure are real
- most serious tasks still need a human in the loop
What I like most here is the organisational angle. Many teams already have automation capability in-house. RPA teams have years of experience designing workflows, governance, and guardrails. They are well placed to lead this shift safely.
AI is now an infrastructure race again
The UK "Tech Prosperity Deal" story is another reminder that this is not just about models.
Microsoft's £22bn investment is part of a wider £31bn package, with new data centres and a supercomputer. Nvidia is talking about the UK as an "AI superpower". Google is committing billions too.
Whether you love the hype or hate it, this is the pattern: countries are treating compute and data centres as strategic infrastructure.
And it raises uncomfortable questions that don't fit into a press release:
- energy use
- public cost
- who benefits and who doesn't
- whether we're heading into a bubble
Link: Read more: UK Tech Prosperity Deal and Microsoft investment
US education leaders are moving fast on policy and procurement
One of the more important signals this week came from the SETDA State EdTech Trends Report.
AI has become the number one US State EdTech priority in 2025, overtaking cybersecurity for the first time.
What's notable is that it's not just talk. States are building real capability:
- hiring AI specialists (Utah, North Carolina)
- publishing classroom guidelines (Massachusetts, Washington, Ohio)
- developing AI literacy training (Wyoming, Massachusetts)
- funding pilots and grants (North Carolina, Virginia)
- procurement guidance focused on safety, interoperability, inclusivity, and evidence
That last part matters a lot. Procurement is where values become real. If you set the standards there, you shape the market.
Link: Download the SETDA 2025 State EdTech Trends Survey and Report
Adoption is diversifying and education is showing up near the top
Anthropic's economic index report had a fun Australia detail: per-capita Claude usage here is much higher than expected.
What stood out to me was the use breakdown. Coding still dominates, but applications are diversifying quickly.
And "educational tutors" sits at number three in Australia.
That lines up with what many of us see on the ground. People are using these tools for learning, not just writing and summarising.
Link: Read the Anthropic Economic Index September 2025 Report
Vibe coding is fast but someone has to clean up the mess
A TechCrunch piece on "vibe coding" captured something I think every leader needs to understand.
AI can accelerate prototyping and scaffolding. But it often creates new work:
- 95% of developers report spending time fixing AI-generated code
- senior developers carry the heaviest load, acting like "AI babysitters"
- issues include hallucinated package names and security vulnerabilities
This is the pattern I keep seeing across fields. AI doesn't remove work. It reshapes it. And it shifts responsibility upward toward review, judgement, and risk management.
That's not a reason to avoid it. It's a reason to budget for oversight and build guardrails.
Link: Read more: Vibe coding has turned senior devs into AI babysitters
A quick schools note from home
Also great to see SA making EdChat available to all secondary students in Term 4. Big fan.
And stepping back, this is the direction of travel: more systems moving from pilots to real access for students.
This is where AI for schools needs to be very practical. Not just "a chatbot exists", but:
- clear classroom norms
- teacher support and training
- procurement standards
- privacy boundaries
- evidence that it improves learning, not just output volume
Where I'm sitting after all this
The headline for me isn't one product.
It's the shape of the next phase:
- AI moving from screens into the environment (glasses, voice, context)
- browsers becoming agents, not pages
- infrastructure and compute becoming national strategy
- education policy catching up through guidelines and procurement
- adoption diversifying, with tutoring and learning near the top
- and a consistent reminder that oversight work doesn't disappear, it changes form
The winners, in schools and everywhere else, will be the teams who keep their eyes on outcomes.
Not theatre. Not dashboards for the sake of dashboards. Real learning, real time saved, real equity, real safety.