Kia ora! Welcome to New Zealand’s weekly roundup of AI news and education.
In case it hasn't crossed your desk yet, Matt Shumer's "Something Big Is Happening" essay did 80 million views last week.
When something hits that wide, it's worth paying attention to. It’s being talked about as one of the most important essays of the year already.
Shumer's core message: the vast majority of knowledge workers have no idea what the current generation of AI tools can actually do. Most stillu use them to rewrite emails and upload documents for analysis. The real shift (delegating entire tasks and getting finished work back) is happening on tools most people have never heard of.
That message is spot on. But what the essay leaves out is the specificity. I sat with it for days, read two strong counterarguments from David Oks and Connor Boyack, and wrote a piece adding context to the conversation.
Please read it, disagree with it, build on it. The thinking only gets sharper when it's challenged.
Happy reading ✌️
Did someone forward you this? Sign up!
academyEX is the EXceptionally proud sponsor of this week's newsletter.
🇳🇿 New Zealand News
NZ sharemarket plunges 2.4% in a single day on AI fears. The S&P/NZX 50 dropped 333 points to 13,198, its lowest since September 2025 and a bigger fall than Trump's "liberation day" tariffs caused last April. Weak Cisco earnings and AI disruption fears drove the sell-off, with Serko down 7.2% in a single session.
2 min read.
Our take: AI disruption is no longer an abstract technology story. It's showing up in KiwiSaver balances and retirement portfolios. The companies that get hit hardest are the ones that sold recurring software subscriptions without a clear AI moat, and NZ has plenty of those on the NZX.
Government moves on AI for breast cancer screening. Health Minister Simeon Brown announced Health NZ will issue a Request for Information inviting organisations with AI image-reading experience to outline how it could support BreastScreen Aotearoa. Around 3,400 women are diagnosed with breast cancer annually. Breast Cancer Foundation NZ called the move "long overdue".
2 min read.
Our take: AI as a second set of eyes for radiologists is one of the most proven use cases globally. NZ's radiology workforce is stretched thin, and this RFI is the first real step from "we should explore AI in healthcare" to actually doing it. The validation process will take time, but this is the right start.
Christchurch AI startup Contented raises $4.1M. The meeting-to-documentation tool closed an oversubscribed seed round led by Altered Capital. Contented captures workplace conversations and uses AI to turn them into compliance checklists, minutes, and statements of advice. The platform supports te reo Maori transcription and has 200+ customers including Synlait, Craigs Investment Partners, and Christchurch Airport. Revenue hit $1.4M ARR before raising any capital.
2 min read.
Our take: Female-founded, bootstrapped to meaningful revenue before raising, oversubscribed round, practical product with real customers. This is what NZ AI success stories look like when substance leads over hype.
Wellington medtech Wellumio raises $7.28M for portable AI stroke scanner. The pre-Series A round, led by Nuance Connected Capital, funds development of Axana, a portable 50kg battery-powered MRI-based brain scanner using AI-enhanced technology for rapid stroke detection within the critical "golden hour." Fewer than 5% of stroke patients currently receive treatment in time.
3 min read.
Our take: This sits at the intersection of AI and medical hardware, which is a much harder category to commoditise than pure software. If Wellumio can validate clinically, the addressable market is enormous: the global stroke diagnostic imaging market exceeds US$8.9 billion.
Political parties reveal positions on AI in election campaigns. A 1News survey found a wide spectrum: Labour won't use AI for image or video generation. National will use AI content with labelling. ACT uses AI for drafting with human oversight. The Greens pledged no generative AI at all. Te Pati Maori noted its MPs have already been targets of AI-fabricated content designed to stoke racial tension.
3 min read.
Our take: Nothing in NZ law currently prevents misleading AI use in political advertising. The Electoral Act 1993 was written for a pre-AI era. Experts are now warning of an "AI arms race" in campaigning, and the first AI-generated election ads have already appeared.
ASB CEO details where the bank is deploying AI. Vittoria Shortt described engineers using generative AI to speed software delivery and AI-enabled call centre workflows that surface information during live calls. ASB is also leveraging group capability from parent Commonwealth Bank of Australia for code translation and customer due diligence.
3 min read.
Our take: A rare on-the-record look at AI inside a major NZ bank. Most bank AI talk stays at the "exploring opportunities" level. Shortt's comments name specific use cases with measurable outcomes, which is a signal that the boardroom-to-production gap is closing at ASB.
Kiwi Chris Liddell joins Anthropic board ahead of blockbuster IPO. The former Carter Holt Harvey CEO, Xero chairman, and Microsoft CFO has been recruited to the board of Anthropic, maker of the Claude AI chatbot. Anthropic recently raised US$30 billion at a US$380 billion valuation. Liddell previously served as deputy Chief of Staff for policy during Trump's first term.
3 min read.
Our take: A Kiwi at the governance table of one of the world's two most valuable AI companies. Liddell's appointment is a practical signal: Anthropic is building the board of a public company, and the IPO runway is shortening.
✨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.
🎙️ The AI Corner Podcast
Join me as I discuss:
Trade Me Property just launched as NZ’s first ChatGPT app. This is a distribution shift from websites to conversations, and it’s going to hit every business.
Live walkthrough of how listings surface mid-chat, why human-written content breaks when agents shop, and why structured data turns into your moat.
App store parallels, OpenAI UX rules, privacy + ads concerns, and why early movers can lock in habits that are hard to unwind.
📚️ AI for Business
Helping leaders and teams adapt, learn, and scale with AI.
1️⃣ Nobody can keep up with AI, and that's not the problem: 43 significant AI releases in January, but only three mattered for client work. Most professionals aren't behind because AI moves fast, they're behind because their strategy is consume more, follow more, try harder. A simple fluency stack to cut the noise and ship one workflow per week.
4-min read.
2️⃣ Output inaccuracies aren't due to hallucinating AI, it's architecture: AI hallucinations cost enterprises $67.4 billion globally in 2024. The difference between unreliable outputs and production-grade work is four layers: Context, Validation, Quality, Routing. Stacked together, 90-96% hallucination rate reduction. Same model, wildly different results.
4-min read.
3️⃣ Knowledge management just shifted from retrieval to computation: 17 files loaded into a single AI session, voice rules, client data, competitive landscape, quality standards. The system computed patterns across all of them simultaneously. Three decades of "second brain" thinking optimised for storage. That assumption just broke.
3-min read.
4️⃣ From discovery call to proposal deck in 20 minutes, the headcount punchline: Same roles, same people, strategy, design, sales, marketing. The shift is compressing the dead time between them so coordination stops turning into days. When elapsed time shrinks, volume doesn't force headcount growth.
3-min read.
5️⃣ AI checks work to 98% accuracy on autopilot: Hallucinated numbers, brand misalignment, chatbot-sounding output. The fix wasn't better prompts, it was moving quality enforcement from human review into code. A Python linter runs deterministic checks, then specialised agents score against 8 quality passes and iterate until every pass locks.
3-min read.
6️⃣ The AI interaction model just changed without many people knowing: Claude asked three questions before writing a single word. Not a bug. Interactive questions mean the model interviews before it executes. Without those questions, generic output in the wrong format. With them, the first draft went straight to production without edits.
2-min read.
7️⃣ Ex-CTO of Australasia's largest venture studio goes solo with AI writing 99.5% of the code: Toby Cox scaled engineering from zero to 70 at Paloma ($250M+ portfolio), then walked away to build Geodde with Claude Code. His take: vibe coding is "hold on for dear life, screwed when it gets hairy". Agentic engineering is the same output with actual understanding underneath.
1-min read.
8️⃣ One customer service ticket cost a business $500k last quarter: Not the ticket itself, but the specialist pulled off a deal to answer it. Customer service latency creates sales latency, but most businesses measure them separately. AI that surfaces answers to frontline teams at the point of need changes the economics entirely.
3-min read.
9️⃣ Who controls the AI lens matters more than Anthropic's Super Bowl ads: Anthropic spent $8 million per spot to frame it as ads or no ads. The reality is messier. Two funding models are emerging: attention (ad-funded, free access) and premium (subscription, gated by income). Both make the same company builder, gatekeeper, and architect of what information flows through. The real question: who controls the lens?
4-min read.
🌍 Tech Updates From Global
The selected top headlines from each major AI tech company.
Anthropic
Closed a US$30 billion Series G at a US$380 billion valuation, the second-largest venture raise in history behind OpenAI's US$40 billion round. Reports annualised revenue of US$14 billion with 500+ enterprise customers spending US$1M+ annually.
Appointed New Zealander Chris Liddell to its board of directors, signalling IPO readiness with an experienced public-company operator.
Claude app shot into the top 10 on app stores after Super Bowl ad campaign mocking AI advertising went viral, driving 11% user growth.
Ongoing fallout from Cowork plugins: software stocks continued bleeding after last week's US$285 billion rout, with global software sector losses now exceeding US$1 trillion.
OpenAI
Released GPT-5.3-Codex-Spark, a streamlined coding model achieving 1,000+ tokens per second on Cerebras wafer-scale chips, the first production deployment of a frontier model on non-NVIDIA hardware.
Launched Frontier enterprise agent management platform, with Uber, State Farm, Intuit, and Oracle as first customers.
Began testing advertisements in ChatGPT for free-tier users in the United States, marking the first step toward ad-supported monetisation.
DeepMind upgraded Gemini 3 Deep Think, a specialised reasoning mode for science and engineering that reportedly solved 18 previously unsolved research problems.
Gemini app surpassed 750 million monthly active users, still trailing ChatGPT's 810 million.
Faces DOJ investigation into whether its deal to power Apple's Siri with Gemini constitutes a monopoly.
Alphabet issued a rare 100-year bond as part of a US$32 billion raise for AI infrastructure.
Microsoft
FTC ramped up antitrust scrutiny, issuing civil investigative demands to Microsoft's competitors over potential anti-competitive bundling of Azure with Copilot.
Continues on pace to exceed US$100 billion capex for FY2026, with power grid constraints limiting data centre expansion.
Publicly framing a move toward AI "self-sufficiency" with in-house foundation models while retaining long access horizons to OpenAI models.
Partnered with Wesfarmers in a multi-year deal expanding AI and cloud services across Bunnings, Kmart, Blackwoods, and Priceline in Australia and NZ.
Meta
Started construction on a US$10 billion, 1 gigawatt AI data centre in Indiana, sized for 2027-2028 operations.
Became the first major tech company to formally tie employee performance reviews to AI tool usage and adoption metrics.
Partnered with ElevenLabs to bring voice AI technology to Instagram and Horizon Worlds.
Amazon
Planning an "AI content marketplace" for publishers to license content for AI training, a direct attempt to industrialise content licensing instead of fighting case-by-case copyright battles.
Custom AI chip revenue exceeded US$10 billion annually with 1.4 million Trainium 2 chips deployed and Trainium 3 launched.
Pledged US$50 billion for AI supercomputing capacity dedicated to US government agencies.
xAI
Lost half of its founding team as six co-founders departed amid the SpaceX mega-merger and Grok controversies.
Musk announced a major restructuring of xAI teams following the departures, suggesting the exits were "push, not pull."
China AI
Multiple Chinese labs released frontier models during Lunar New Year: ByteDance launched Doubao 2.0 matching GPT-5.2 capabilities, Alibaba unveiled RynnBrain for robotics, Zhipu AI released GLM-5 with long-running agent capability, and MiniMax shipped M2.5.
ByteDance's Seedance 2.0 video AI triggered a Hollywood copyright crisis after viral AI-generated videos featuring recognisable actors prompted Disney to issue a cease-and-desist letter.
AI Safety
An unprecedented wave of safety researcher departures hit OpenAI, Anthropic, and xAI simultaneously. Multiple departing researchers publicly warned about rushed development and downplayed risks, with one Anthropic safety chief describing the world as "in peril."
Salesforce reportedly cut fewer than 1,000 roles including in its Agentforce AI agent product team, another signal that "AI efficiency" is being operationalised via headcount decisions.
👋 Mike & Erin