Kia ora! Welcome to New Zealand’s weekly roundup of AI news and education.
We’re opening the door a little wider at The AI Corner.
Over time, The AI Corner has become a place to share useful signal on what is happening in AI and where things are heading in New Zealand.
As it grows, we want it to become more valuable than just one person’s curation. We want it to be a stronger hub for the AI ecosystem here. One that surfaces more smart people, more practical work, and more innovation.
A big part of that is helping more Kiwi AI innovators reach a wider audience than they may be able to on their own.
That does not mean we’re opening the floodgates. Nor does The AI Corner become just a promotional channel for businesses. You’ve trusted us to deliver signal through the noise, and that stays the priority.
What we’re interested in is hearing from consistent creators already contributing to the ecosystem who may want to collaborate with The AI Corner.
That could be through:
content and commentary
events and meetups
workshops and education
tools and resources
research and reports
practical AI use cases inside teams and businesses
What matters to us:
you create consistently
you bring a unique, real point of view
you want to contribute to the ecosystem, not just promote yourself
To repeat, this is not an open publishing platform or a free-for-all for self-promo. Everything will be curated and filtered through an editorial lens.
If that sounds like you, reply with a short note on who you are, what you create, and why you’d be a strong fit for The AI Corner.
Happy reading ✌️
Did someone forward you this? Sign up!
🇳🇿 New Zealand News
Xero and Anthropic announced a multi-year partnership to integrate Claude AI into Xero's platform, powering the AI assistant JAX to automate cashflow tracking, invoice chasing and financial reporting. Xero financial data will be available inside Claude, but data shared between platforms is used only for specific sessions and never used to train Claude's models. 2 min read.
Our take: Anthropic opened a Sydney office three weeks ago. Now they have signed Xero, the platform through which millions of small businesses operate. This is not them exploring NZ; it’s them locking up distribution channels before anyone else moves. Also "never used to train Claude's models" is the sentence that makes this deal possible. Accountants operate in one of the most regulated professions in NZ. If financial data leaked into training sets, the partnership collapses overnight.
Health NZ warned mental health staff in Rotorua Lakes district that using ChatGPT, Claude or Gemini for clinical notes could result in formal disciplinary action. The same day, the Public Service Association hit back, arguing a threatening memo is not a training programme. PSA National Secretary Fleur Fitzsimons said clinical staff are turning to free AI tools because of "enormous pressure" on workloads and called on Health NZ to invest in approved tools instead. 3 min read.
Our take: The PSA's response is the more interesting half. Fitzsimons' line: "Threats will not fix Health NZ's AI problem, investment will". A union making a pro-AI adoption argument. The traditional script: management deploys AI, union resists. This is the opposite. Workers adopted AI on their own, management is trying to stop them, and the union is saying let them use it properly.
NZ agtech company Halter closed a US$220M (NZ$377M) Series E led by Peter Thiel's Founders Fund, doubling its valuation to US$2B in nine months. The company has sold 1 million AI-powered solar cattle collars to 2,000+ farmers across NZ, Australia and the US. Capital will fund expansion into Ireland and the UK, plus 200 new hires at NZ headquarters. Founded in 2016 by Craig Piggott, now 31. 3 min read.
Our take: Thiel's Founders Fund leading the round is a signal to every Silicon Valley fund that NZ primary sector AI is investable at scale. The next Kiwi agtech founder pitching in San Francisco no longer has to explain why farming matters to tech investors. Halter already made that argument at $2B.
BusinessDesk asked 17 government agencies to disclose AI and automation spending since January 2023, and 13 responded, revealing at least $20.2m in disclosed spending. The true figure is likely considerably higher. A prototype AI assurance regime is expected to be ready for testing with agencies by end of June 2026. 2 min read.
Our take: $20.2m across 13 agencies since 2023 is not a large number. Compare it to the $13B total government technology spend: roughly 0.15%. Not spending too much on AI. Spending too little, too slowly, without the governance structures that would make larger spending safe.
A Kapiti teenager applying for a supermarket job had his personality judged by Sapia AI, which told him his "self-belief" could "alienate" people. The AI told another dyslexic applicant he was "not one to try new things easily". Over a million job applicants across Australia and NZ have been screened by the tool. More applicants came forward with "laughable" experiences of being judged and rejected by AI personality assessments. 3 min read.
Our take: A 16-year-old applying for a checkout job was told by an algorithm his self-belief could alienate people. Personality analysis for a role that requires scanning barcodes. The mismatch between the tool's sophistication and the job's complexity is the story.
📚️ Mike’s Takes From The Week
Helping leaders and teams adapt, learn, and scale with AI.
1️⃣ Jason Lemkin's hiring test for AI fluency is one question: What commercial AI tool has this person brought into the organisation this month? Not played with or read about, but actually brought in. The principle is sound: people who move organisations forward spot better ways of operating and know how to apply new capability practically, not pile tool after tool into the business for the sake of it.
2-min watch.
2️⃣ Most people use Claude like a search engine with better grammar. Time to build a Chief of Staff: Three layers make it work: connectors plug into Gmail, Slack, and HubSpot; skills encode how the work gets done; context files give the model enough background to make decisions worth trusting. No code, just plain-text files and a web interface. The compounding starts when morning briefings, pipeline reports, and transcript processing all run without anyone at a desk.
5-min read.
3️⃣ Most AI agents in NZ businesses are built on an island. Islands do not compound: Australian businesses are connecting the layer underneath so the next agent costs a fraction of the first. In NZ, scattered AI projects all work fine on their own but share nothing: a signal captured in customer service never reaches the sales team, and agent number ten costs the same as agent number one. The fix is shared infrastructure, standardised access points and a single governance layer, so deploying a new agent becomes a configuration change rather than a development project.
5-min read.
4️⃣ 61% of Australian consumers dislike or hate agentic commerce. 85% of businesses are building for it anyway: Australia Post's annual eCommerce report shows $82.6 billion in online spend and 14% growth, but the consumer trust gap is the real story. The categories people would consider delegating to an agent are groceries, clothing, and books: repeat purchases, commoditised options, low emotional stakes.
2-min read.
5️⃣ Business book frameworks gathering dust on a bookshelf is a solved problem: Feed them into Claude Code as individual skills that apply the framework against live scenarios, not as summaries. Different books start interacting with each other: a coaching diagnostic feeds into an offer design, which feeds into a sales conversation structure. Each book added makes the system more useful across every task.
1-min watch.
🌍 Tech Updates From Global
The selected top headlines from each major AI tech company.
OpenAI
Fidji Simo declared "Code Red" at all-hands, pivoting away from experimental products to focus on coding tools and enterprise after OpenAI's enterprise share fell from 50% (2023) to 27% while Anthropic captured ~40%.
The clearest admission yet that Anthropic is winning the enterprise race.Shut down the Sora video app after $15M/day inference costs and only $2.1M lifetime revenue; Disney's $1B deal collapsed simultaneously.
First major product kill signals capital discipline ahead of the IPO.Completed pre-training of next-gen model "Spud"; Altman told employees it could "really accelerate the economy," with release expected within weeks.
If Spud is GPT-6, the competitive landscape resets again.Funding round reached $120B ($850B valuation); SoftBank secured $40B unsecured 12-month loan implying lenders expect IPO by Q4 2026.
The largest private funding round in history keeps getting larger.Launched 20+ Codex Plugins (Slack, Figma, Notion, Cloudflare, Gmail, Google Drive) expanding Codex into an integrated work platform.
Codex is becoming an operating system for developer workflows, not just a code assistant.Court ordered 20M ChatGPT logs produced to copyright plaintiffs across 16 consolidated lawsuits.
The most significant discovery ruling in AI copyright law to date.
Anthropic
Judge Lin issued preliminary injunction against Pentagon blacklisting, calling it "classic First Amendment retaliation".
Strongest judicial pushback yet against politically-driven AI procurement.Leaked data store revealed "Claude Mythos" model described as a "step change" in capabilities; Bloomberg reports October 2026 IPO consideration.
Accidental disclosure confirms something significant in the pipeline; two AI labs racing to Q4 2026 IPOs.Revenue run rate hit ~$19B (80% enterprise); Claude Code alone at $2.5B ARR; 20% of US companies now pay for Anthropic tools.
Fastest product ramp in B2B software history.Claude Code released Auto Mode, Telegram/Discord Channels, and computer use via Claude Cowork for Pro and Max users.
Three product launches in one week expanding Claude Code from IDE tool to cross-platform agentic system.MCP crossed 97M monthly SDK downloads (up from 2M at Nov 2024 launch); signed $200M multi-year Snowflake partnership.
MCP is now the de facto standard for connecting AI to external tools.
Launched "switching tools" letting users transfer full chat histories and memories from ChatGPT and Claude into Gemini.
Lowering switching costs is how platform wars are won.Released Gemini 3.1 Flash Live, a native audio-to-audio voice model supporting 90+ languages with lower latency.
Real-time multimodal voice is the interface layer for agentic AI.Google Search Live expanded to 200+ countries with camera-as-search; Gemini TV features coming to NZ this spring.
Camera-as-search is Google's most differentiated consumer AI feature.Released Lyria 3 Pro music generation producing 3-minute structured tracks; Gemini Personal Intelligence rolling out free to all US users.
Two product launches expanding Gemini's consumer and creative footprint.
Microsoft
Copilot has only 6M daily active users vs ChatGPT's 440M; just 3.3% of M365 users who try Copilot actually pay for it.
The adoption gap between free usage and paid conversion remains the central AI business challenge.Stock fell 24% YTD (worst quarter since 2008); AI capex projected at ~$146B for FY2026 against $13B AI revenue run rate.
Market pricing in the possibility that AI spending doesn't generate returns fast enough.GitHub Copilot will train on user interaction data by default from April 24; users must manually opt out.
Most significant AI training data policy change since Microsoft acquired GitHub.
xAI
Baltimore became first US city to sue xAI over Grok deepfakes; Dutch court issued Europe's first deepfake ban with EUR 100K/day fines.
Municipal lawsuits and European court orders represent new legal fronts.Nine of 11 original co-founders have left; Musk acknowledged "xAI was not built right first time around."
Co-founder retention below 20% is a serious organisational signal.Announced Terafab, a $25B chip fab targeting 2nm process nodes, jointly backed by Tesla, SpaceX, and xAI.
Ambitious but Bloomberg questions the timeline and cost viability.
Perplexity
Samsung launched its Windows browser with Perplexity-powered "Browsing Assist," reaching over 1 billion Samsung devices.
Distribution through Samsung's device base is Perplexity's biggest partnership to date.Launched Market Research Agent with premium data from CB Insights, PitchBook, and Statista for Pro/Max subscribers.
Specialised vertical agents with premium data differentiate from generic AI search.
Amazon
Spending an expected $200B on AI infrastructure in 2026 while cutting ~30,000 jobs; Gartner classifies Amazon as a "digital nation state."
$200B in AI spend from a single company is without historical precedent.Eight validated attack vectors disclosed in AWS Bedrock spanning agent hijacking, prompt poisoning, and data exfiltration.
Enterprise AI security is now a live operational concern, not theoretical.
Apple
Bloomberg: Apple will open Siri to Claude, Gemini, and other AI models via "Extensions" in iOS 27.
Multi-model Siri transforms Apple from AI laggard to distribution layer for all major providers.Testing standalone Siri app with chat interface; WWDC 2026 announced for June 8-12 with explicit AI focus.
Most significant Siri redesign since 2011.
Meta
Increased El Paso data centre investment from $1.5B to $10B; cut 700 more employees while approving $921M in executive stock compensation.
Escalating capex alongside layoffs and exec payouts will fuel scrutiny.Testing AI-powered shopping on Instagram/Facebook; expanding affiliate partnerships with Amazon, eBay, Temu, and others.
AI-enhanced ad engagement is Meta's most direct path to AI-driven revenue.
NVIDIA
Shifting from fabless chip designer to turnkey supplier of entire AI data centre infrastructure (seven chips, five-rack systems).
The pivot from component supplier to full-stack infrastructure company is now the dominant NVIDIA narrative.AWS will deploy 1M+ NVIDIA GPUs in 2026; Microsoft disclosed hundreds of thousands of Grace Blackwell GPUs already deployed.
Hyperscaler NVIDIA deployments are accelerating, not plateauing.
✨A few people have asked…
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.
👋 Mike & Erin