Kia ora! Welcome to New Zealand’s weekly roundup of AI news and education.
Anthropic is on a tear. From putting Claude Code into the hands of non-technical knowledge workers, to refusing the Pentagon and hitting #1 on the App Store, to landing inside Microsoft 365 Copilot this week. Three weeks, three moves that rewrite how Claude gets distributed. And now they have named NZ a priority market, with an exec team visiting end of March.
More from a Caffeine Daily article in the NZ News section.
Additionally, we’re in the habit of finding and sharing our favourite AI meme video from the net each week. Here’s this weeks:
Happy reading ✌️
Did someone forward you this? Sign up!
🇳🇿 New Zealand News
One of the big three AI labs just set up shop across the Tasman. Anthropic opened a Sydney office and named NZ a priority, citing 8th-highest Claude usage per capita globally. The company is exploring local compute to address data residency, and its exec team visits NZ end of March to meet policymakers.
3 min read. Via Caffeine Daily.
Our take: NZ ranked 8th per capita for Claude usage and we still don't have a single Tier-1 AI lab with a local presence. Anthropic coming down under won't fix that overnight, but it is the first time a frontier lab has named NZ by name rather than lumping us into "Asia-Pacific".
InternetNZ's annual survey found 47% of New Zealanders are concerned about AI while just 12% are excited, and NZ ranks third-to-last globally in AI trust across 47 countries. 64% believe government regulation is inadequate. University of Canterbury academic Olivia Erdelyi, who sits on the international AI standards committee, says legal uncertainty is "actually inhibiting innovation", not protecting it. 4 min read.
Our take: Third-to-last in AI trust out of 47 countries. That is worse than most of Europe, worse than most of Asia, and worse than Australia. The countries that rank higher in trust are, by and large, the ones with clearer regulation. This correlation is not a coincidence, and our country’s leaders are literally watching a competitive advantage (well, more like a parity necessity) become a weakness.
Only 41% of Kiwi workers use AI, compared to 91% in India, and NZ's new $3.5B AI factory will pipe its output offshore. Tech expert Mark Laurence told RNZ that NZ is "still a nation using AI to change the tone of an email" while the world redesigns entire workflows. He wants national capability training programmes to go hand-in-hand with infrastructure investment. 3 min read.
Our take: Laurence keeps saying the same thing because nobody is doing anything about it. This is his third major media appearance in three weeks making the same argument. The diagnosis is consistent. The prescription (national training programmes) is clear. The government's response is to build a data centre that sends its output to Australia.
A free AI literacy programme for Years 5-10, adapted from MIT and incorporating te ao Maori perspectives, rolled out nationally on 9 March with the goal of reaching every NZ school by the end of Term 1. Piloted across ten schools in Term 4 2025 and evaluated by NZCER, the programme offers flexible delivery (full day or eight 30-minute blocks). RNZ featured a Westlake Girls teacher discussing the rollout. 2 min read.
Our take: The programme is good. The ambition (every school by end of Term 1) is great. The question is whether "one day of AI" becomes a meaningful shift in how schools teach critical thinking about technology, or a box-ticking exercise that lets schools claim they've covered AI without changing anything else in the curriculum.
Datagrid secured full consent to build a $3.5B, 280MW AI data centre on a 49-hectare site in Makarewa, making it NZ's second-largest electricity user after Tiwai Point. The 78,000 sqm campus will host high-density GPU clusters for global AI and cloud providers, connected to Transpower's network via a dedicated 220kV substation. A trans-Tasman subsea cable landing at Oreti Beach, the South Island's first international fibre link, was approved simultaneously. The Tasman Ring Network will connect Invercargill directly to Sydney and Melbourne, with a domestic route bypassing the Wellington fault zone. Construction creates 1,200+ jobs. But Datagrid's output gets piped offshore, not to NZ businesses, and the company has not disclosed its customers. 3 min read.
Our take: NZ is building AI infrastructure that serves overseas markets, powered by NZ's renewable electricity, on NZ's land, near NZ's cultural sites. The construction jobs will benefit locally also. The long-term question is whether Southland becomes a digital export economy or an electricity supplier with a fancy building.
academyEX is the EXceptionally proud sponsor of this week's newsletter.
📚️ AI for Business
Helping leaders and teams adapt, learn, and scale with AI.
1️⃣ Claude Code Use Case: One transcript, 14 documents, five live systems, 11 minutes: A Claude Code command turned a single call transcript into file notes, task lists, compliance logs, Gmail drafts, a HubSpot deal, a calendar event, and a Webflow blog post. The point isn't the demo. It's that most teams still use humans as middleware, translating between systems one tab at a time.
3-min watch.
2️⃣ Microsoft just shipped its most important product, and it's built on Anthropic: Copilot Cowork launched with AI agents that plan and execute across the full M365 stack. After $13B into OpenAI, Microsoft invested $5B directly into Anthropic and got $30B in Azure compute back. Not a hedge, empire building.
2-min watch.
3️⃣ Agentic commerce didn't fail. OpenAI just built it in the wrong order: Instant Checkout got pulled back, and the market assumed consumer demand wasn't there. The behaviour is heading that way, but AI is reshaping discovery while checkout rails, incentives, and merchant systems still sit in someone else's hands.
6-min read.
4️⃣ The data leader who skipped Copilot on purpose and went straight to the P&L: Sean Narayan spent two years building AI at Bidfood NZ targeting inventory management, order processing, and margin improvement. $800 per user across 1,000 users with unproven ROI versus AI that changes what gets ordered and when. The maths made the decision for him.
45-min listen.
5️⃣ The CLAUDE.md file is the receptionist in the lobby: Before a single word is typed, Claude Code reads a markdown file that shapes every output and every routing decision. Three index files, three hops. One sentence in the lobby gets to a fully contextual output. No index files and the AI just wanders hallways dropping files at random.
3-min read.
6️⃣ The Pentagon asked. Anthropic said no. OpenAI said yes: Claude hit number one on the App Store the week Anthropic was banned. ChatGPT uninstalls surged 295% in a single day. The consumer market rewarded principle. The contract market rewarded flexibility. Both signals will shape how this industry develops.
6-min read.
🔥 Claude Cowork Workshop.
3 hours. No code. You walk away knowing how to build anything with AI.
No technical expertise needed, this is a hands-on Claude Cowork workshop to learn the key tools and methods to build anything, websites, applications, automations, and more.
31st March - Auckland
Limited Tickets - $199 +GST
Use Waboom200 for $200 off
Become the most powerful person in the room.
🎙️ The AI Corner Podcast
This week's guest is Fiona Millar, Digital Transformation Manager of Christchurch International Airport. Hear:
Why the airport chose ChatGPT over Copilot after running controlled cohorts, and how a self-taught specialist flipped the script.
How embedding an AI automation specialist into teams for three-week sprints unlocked 14 use cases in a single round.
What happens when you put your entire executive team through 12 weeks of mandatory AI training, CEO included.
✨A few people have been asking…
It’s Mike here, I run The AI Corner.
I’m not just into writing about AI. I help businesses grow without adding headcount by shipping AI systems that cut costs and increase throughput.
🌍 Tech Updates From Global
The selected top headlines from each major AI tech company.
OpenAI
GPT-5.1 models retired, auto-migrating all conversations to GPT-5.3 or GPT-5.4 variants. Aggressive deprecation cycle compresses the model zoo.
Google and Microsoft app integrations in ChatGPT now support write actions: drafting emails, creating docs, scheduling meetings. ChatGPT becomes an action layer across competing ecosystems.
Robotics hardware lead Caitlyn Kelanowski resigned over the Pentagon deal, citing governance concerns. Internal fractures over military AI are now costing OpenAI senior talent.
Oracle and OpenAI scrapped plans to expand the flagship Stargate data centre in Texas; existing 0.5GW facility still on track. Scaling back signals capital discipline, not retreat.
ChatGPT launched interactive visual learning modules for 70+ maths and science topics with real-time formula manipulation. Education use case getting serious product investment.
Workspace analytics launched for Enterprise and Edu with benchmarks, impact surveys, and a new analytics viewer role. Enterprise buyers now get usage data to justify spend.
Codex app update (26.312) shipped theming, revamped automations with local/worktree execution, and custom reasoning levels. Developer tooling getting rapid iteration.
Anthropic
Claude integrated into Microsoft 365 Copilot via the Frontier programme; Anthropic's Cowork technology powers Copilot Cowork. Claude inside Microsoft's enterprise suite is a distribution breakthrough.
Claude now creates interactive charts, diagrams, and inline visualisations in responses. Closing the gap with ChatGPT's visual output capabilities.
Claude for Excel and PowerPoint add-ins updated with shared conversation context, skills support, and LLM gateway for Bedrock/Vertex/Foundry. Multi-cloud model routing inside Office apps.
1M token context window now GA for Opus 4.6 and Sonnet 4.6 at standard pricing; media limit raised to 600 per request. Context length is no longer a beta feature or premium add-on.
Named Sydney as Asia-Pacific headquarters. APAC expansion with a regulatory-friendly jurisdiction.
Sued the Trump administration over supply-chain risk designation, seeking immediate injunctive relief after federal agencies began dropping Claude. Legal escalation turns the Pentagon split into a constitutional dispute.
Invested $100M in the Claude Partner Network to expand the ecosystem. Ecosystem investment signals platform ambitions beyond the model layer.
Claude Code multi-agent Code Review system launched for pull request reviews. Agentic code review directly competes with GitHub Copilot's rebuilt review.
Released Gemini Embedding 2 Preview, the first multimodal embedding model supporting text, image, video, audio, and PDF in a unified space. Multimodal embeddings unlock cross-format search and retrieval at scale.
Major Gemini Workspace update: AI-powered drafting, editing, and search across Docs, Sheets, Slides, and Drive for Pro/Ultra subscribers. Google's enterprise AI suite catching up to Microsoft's Copilot integration depth.
Vertex AI partner model evaluations now support Anthropic and Llama models in the Gen AI evaluation service. Google opening its evaluation infrastructure to competitors is a platform play.
DeepMind published "The Abstraction Fallacy: Why AI Can Simulate But Not Instantiate Consciousness". Google positioning itself at the frontier of AI philosophy, not just benchmarks.
Microsoft
Copilot Wave 3 launched with Anthropic Claude models, Copilot Cowork (built on Anthropic), and Work IQ intelligence layer. The biggest M365 AI update yet, and it runs on a competitor's model.
Microsoft 365 E7 "Frontier Suite" announced at $99/user/month, bundling E5 + Copilot + Agent 365. New top-tier pricing sets the enterprise AI ceiling.
Agent 365 agentic control plane GA date set for May 1 at $15/user/month. Enterprise agent orchestration getting a firm ship date.
Copilot Health launched in the US with personalised health advice using medical records, wearables from 50+ devices, and data from 50,000+ hospitals. Healthcare is Microsoft's AI vertical beachhead.
GitHub Copilot for Azure reached GA with Agent Mode for agentic DevOps workflows in VS Code. DevOps automation now generally available.
xAI
xAI US market share grew from 1.9% (Jan 2025) to 17.8% (Jan 2026) per Click-Vision stats. Fastest market share growth in the industry, regardless of controversy.
Perplexity
Held inaugural developer conference "Ask" in San Francisco, unveiling "Personal Computer", a local AI agent for Mac with mandatory user confirmation and audit trail. Perplexity pivoting from cloud to on-device, positioning against OpenClaw on security.
Ordered to stop deploying shopping AI agents on Amazon. First major legal boundary drawn around agentic commerce.
Amazon
Bedrock AgentCore Runtime added stateful MCP server features: elicitation, sampling, real-time progress, and session context. MCP support in Bedrock is infrastructure validation for the protocol.
Amazon Connect launched AI-powered manager assistant querying 150+ metrics via natural language. Contact centre AI moving from agent-facing to manager-facing.
Meta
Yann LeCun (who recently left Meta) raised $1.03B for AMI Labs to build "world models", backed by Bezos, NVIDIA, and Eric Schmidt.
Planning four new in-house AI chips (MTIA 300-500) by end of 2027; MTIA 300 already in production.
Reportedly planning layoffs affecting up to 20% of workforce amid mounting AI infrastructure costs.
NVIDIA
Reportedly planning NemoClaw open-source AI agent platform for enterprises, pitched to Salesforce, Cisco, and Google.
Samsung
Galaxy S26 series shipped March 11 with agentic Gemini 3 AI, Privacy Display, AI call screening, Perplexity integration, and natural-language photo editing.
👋 Mike & Erin