In partnership with
Kia ora! Welcome to New Zealand’s weekly roundup of AI news and education.
Can you help us crack 1,000?
Last year, more than 100 of you filled out our reader survey. This year, we’re aiming higher…
Our annual reader check-in takes 2 minutes and helps us understand who’s reading, what you want more of, and what we should stop wasting your time with.
That could mean better AI news, more local examples, new useful resources, and fewer bits that don’t belong here.
Reminder: We’re hiring a Lead AI Engineer at Allexive. DM me or apply here.
Happy reading ✌️
Did someone forward you this? Sign up!
🇳🇿 New Zealand News
The Ipsos AI Monitor 2026, a global survey of AI attitudes across 32 countries, found 65% of New Zealanders feel nervous about products and services that use AI, down only one percentage point from 66% the previous year, while just 40% feel excited about AI, down from 44%. New Zealand sits in a cluster of Anglosphere countries (Australia, Canada, the US, Great Britain, Ireland) where people are more nervous than excited. Only 21% of Kiwis trust AI outputs enough not to double-check them. Separately, 85% say AI use in products and services should be disclosed.
3 min read
Our take: The Anglosphere clustering is worth examining because it groups NZ with markets that share broadly similar regulatory traditions, media environments, and cultural attitudes toward institutional trust, which means the nervousness is not simply a function of less AI exposure or lower digital literacy. Countries in the Global South and parts of Asia where AI adoption is more clearly associated with economic uplift tend to show more enthusiasm, which suggests the NZ scepticism is partly a product of experiencing AI as a threat to existing employment and information environments rather than as an expansion or growth opportunity.
London-based Odyssey, co-founded by Auckland University graduate Jeff Hawke, closed a US$310m (NZ$529m) Series B at a US$1.45 billion valuation backed by Amazon, GV, AMD Ventures, and Icehouse Ventures, building AI world models that learn from real-world simulation. In the same week, New York startup Sea12, co-founded by NZ's Thomas Mosen, announced a US$25m Series A from Caffeinated Capital for its back-office AI platform serving steel, construction, and logistics firms. Sea12 started with $30,000 and two interns in May 2024 and now runs seven-figure annual revenue with 14 staff.
3 min read - Odyssey, 3 min read - Sea12
Our take: The two raises reflect fundamentally different bets on where AI value accrues. Odyssey is betting on a new architectural layer, world models trained on simulation rather than text, where the commercial upside is in robotics, logistics, and physical infrastructure but so is the capital intensity. Sea12 is betting on workflow specificity, that steel companies and construction firms will pay to automate the document chaos that generic AI tools cannot reach without deep vertical integration. Both theses are credible; they are just measuring success on completely different timescales.
Research from Sharesies, which has more than one million users across New Zealand and Australia, finds one in three Kiwis now turns to AI tools for investment advice. Financial experts quoted in the piece warn that AI investment tools may reflect training biases, cannot account for individual tax situations, and one analyst described the dynamic as "sitting across from a gambler who happens to know a lot of financial history" when the tool lacks real-time market data.
3 min read
Our take: The Financial Markets Authority licenses financial advisers precisely because investment advice given without understanding a client's full situation produces foreseeable harm at scale, and nothing in the current regulatory framework applies that standard to an AI that gives the same output to a 22-year-old with $5,000 and a 55-year-old with a mortgage. The FMA has been watching AI in financial services, but one-in-three usage at a platform the size of Sharesies is the kind of scale that makes "watching" an insufficient response.
APRA AMCOS confirmed that works by New Zealand artists including Lorde, Six60, Bic Runga, Dave Dobbyn, Marlon Williams, Stan Walker, Split Enz, Aldous Harding, Kaylee Bell, and Ché Fu appear in four AI training datasets shared across the AI development community without consent, licence, or payment. The allegation follows an investigation by The Atlantic, which built an AI Watchdog tool allowing anyone to search which artists' songs appear in those datasets. APRA AMCOS projects ANZ creators will lose more than NZ$500 million over four years without a mandatory licensing framework.
3 min read
Our take: The Atlantic's Watchdog tool changes the dynamics of this debate in a specific way, because until now the question of which songs were in which datasets was opaque and disputable, artists could allege but not confirm. A searchable tool that surfaces specific songs from specific artists in specific datasets shifts this from a theoretical harm to a documented one, and that changes what both courts and regulators can do with the information compared to abstract copyright claims.
A 1News social experiment put a group of Auckland University of Technology students in direct conversation with ChatGPT on topics including deepfakes, political manipulation, job losses, and AI therapy companions, and they came away with more anxiety, not less. Senior lecturer in AI at Victoria University of Wellington Andrew Lensen said he was "constantly frustrated" that a few large US tech companies control the future of AI while NZ political leaders are "not confronting that". In response to 1News, Minister of Science, Innovation and Technology Penny Simmonds described the Government's approach as "light-touch, proportionate and risk-based, relying primarily on existing, technology-neutral laws".
4 min read
Our take: Minister Simmonds' response frames "building public trust" as a goal of the Government's approach, but the Ipsos AI Monitor published two days earlier found only 40% of Kiwis feel excited about AI and 65% remain nervous, down only marginally from a year ago despite significant public investment in AI promotion. The question is whether the current approach is actually building that trust or simply not yet losing it, which is a different benchmark.
The best prompt engineers aren't typing. They're talking.
Power users figured this out early: speaking a prompt gives you 10x more context in half the time. You include the edge cases, the examples, the tone you want — because talking is fast enough that you don't skip them.
Wispr Flow captures everything you say and turns it into clean, structured text for any AI tool. Speak messy. Get polished input. Paste into ChatGPT, Claude, Cursor, or wherever you work.
89% of messages sent with zero edits. 4x faster than typing. Works system-wide on Mac, Windows, and iPhone.
📚️ Mike’s Takes From The Week
Helping leaders and teams adapt, learn, and scale with AI.
1️⃣ Satya Nadella's 65M-view essay hides a five-point competitive checklist: Choosing which platform to use is not the advantage, Nadella argues. The businesses pulling ahead load proprietary knowledge in: best calls, win/loss reasons, decisions, outcomes, building an asset no competitor can replicate and no model change can erase.
3-min read.
2️⃣ Trade Me ran one training day and adoption shot up. Eight lessons from what went wrong first: Loren Thomas at Trade Me shares the wins worth copying and the failures worth avoiding. Too slow on rollout, no data governance at launch, measuring login counts instead of outcomes. One NZ business with 1,000-plus staff.
3-min listen.
3️⃣ SpaceX's IPO filing is an AI story, not a rocket story: The $28.5T total addressable market breaks down as $26.5T AI, $1.6T connectivity, $370B space. The Cursor acquisition at $60B gives SpaceX a product developers use daily and a feedback loop on every line of code.
3-min read.
4️⃣ Creating The AI Corner newsletter has dropped from an 8 hour job to a 45 minute job: Starting from a van in South America 18 months ago, each version now captures what worked and what to adjust, so the system does not just produce the next newsletter but improves it.
3-min read.
🛠 Latest Finds From The Web
Helping advanced builders stay at the frontier of AI.
1️⃣ Breaking down Loop Engineering. LangChain's Sydney Runkle breaks the agent harness into four nested loops. Loops one and two are where most teams still live. The value compounds in three and four.
Article
2️⃣ Anything you can benchmark with AI is already a commodity. Sarah Guo's "The Untrainable" is the clearest articulation of where durable value sits in the AI game: work whose correctness is private, expensive to verify, and locked behind permission. The MIT stat is the concrete anchor. Coding agents lifted code written by 180% but code shipped by only 30%. Intelligence got cheap. Everything else still runs through a person.
Article
🌍 Tech Updates From Global
The selected top headlines from each major AI tech company.
OpenAI
OpenAI launched its Partner Network with Accenture, BCG, McKinsey, Bain, and PwC as founding partners, backed by $150M, targeting 300,000 certified consultants by end of 2026. (Jun 16)
Enterprise admins gained a Global Admin Console with per-user, per-product, and per-model credit tracking, group-level spend caps, and a unified Cost API. (Jun 18)
GPT-5.6 described as a "meaningful improvement" by OpenAI's chief scientist, with a rumoured 1.5M token context window; no official release date confirmed within the window. (Jun 16)
Anthropic
Commerce Secretary Lutnick threatened criminal and civil penalties unless Anthropic obtained US government approval before granting foreign nationals access to Fable 5 and Mythos 5. (Jun 14)
Trump told Axios he no longer views Anthropic as a national security threat; Anthropic's international MD said model access would be restored "in the coming days" from Seoul. (Jun 19)
Amodei and DeepMind's Hassabis joined G7 leaders at Évian-les-Bains, calling for a US-led coalition on frontier model standards, chip trade excluding China, and AI-enabled bioterrorism defences. (Jun 17)
Claude Design shipped a major overhaul: importable GitHub design systems, live WYSIWYG canvas editing, two-way /design-sync with Claude Code, and PDF/PPTX export. (Jun 17)
Noam Shazeer, Gemini co-lead and "Attention Is All You Need" co-author, is joining OpenAI, less than two years after Google paid a reported $2.7B to bring him back via the Character.AI deal. (Jun 18)
John Jumper, 2024 Nobel Chemistry laureate and AlphaFold co-creator, announced his departure from DeepMind for Anthropic in the same week as the Shazeer departure. (Jun 19)
Google Vids AI avatars opened to free personal accounts (10 videos/month), expanded presets from 23 to 53, and added 16 new languages. (Jun 17)
Microsoft
Copilot Cowork reached general availability worldwide, priced at $0.01 per Copilot Credit, with Databricks Genie, Dynamics 365 CE plugins, image generation, and partner plugins at launch. (Jun 16)
Work IQ APIs reached GA, giving third-party agents programmatic access to M365 email, calendar, files, and people signals via A2A, remote MCP, and REST under Copilot Credits pricing. (Jun 16)
SharePoint shipped custom AI skills, letting organisations teach Copilot to perform repeatable tasks in a consistent house style across the tenant. (date unconfirmed, rolling out Jun 2026)
xAI
Grok Imagine Video 1.5 reached full GA on web, iOS, and Android, generating 6-second 720p clips with synced audio in ~25 seconds at $4.20/minute, 86% below Sora 2 Pro. (Jun 16)
A Pentagon court declaration revealed Grok was used to identify and strike more than 2,000 targets within 96 hours during the Iran conflict, the first public confirmation of Grok's direct military targeting use. (Jun 16)
Amazon
AWS Summit New York (Jun 17) unveiled a coordinated agentic AI stack: Kiro agentic IDE with a Pro Max tier and iOS app, Bedrock Managed Knowledge Base (GA), Web Search on Bedrock AgentCore (GA), AWS Context knowledge graph, AWS Continuum AI security platform, and DevOps Agent with release management capability.
Apple
Tim Cook warned in a WSJ interview that AI hyperscalers have triggered a global memory chip shortage (nicknamed "RAMageddon"), fourfold chip cost increases, and unavoidable iPhone, Mac, and iPad price hikes. (Jun 17)
Siri AI is blocked in China pending Cyberspace Administration approval of the underlying Gemini model, with Alibaba reportedly lined up as a partner for Chinese users. (Jun 10)
Meta
Meta launched "AI Mode" on Facebook, a plain-language search powered by Muse Spark that synthesises answers from public Groups and Reels, alongside AI video editing and a "Wear It" virtual try-on feature for sports jerseys. (Jun 15)
Over 1,600 Meta engineers signed a petition against keystroke and screen-tracking for AI training data collection; the Applied AI unit is described internally as a "gulag" by some employees. (Jun 17)
NVIDIA
NVIDIA released GR00T-H-N1.7, a 3B-parameter AI foundation model for surgical and healthcare robotics, as the first commercially licensed AI model specifically for that use case. (Jun 17)
Perplexity
Perplexity launched Brain, a self-improving agent memory system for its Computer platform, building a living context graph from completed work that self-teaches overnight, improving correctness by 25% and cutting per-task cost by 13%. (Jun 18)
Mistral
France's Prime Minister announced the civil service would shift to Mistral AI following the US export controls on Anthropic's Fable 5, positioning Mistral as a sovereignty alternative. (Jun 16)
✨A few people have asked…
It’s Mike here, I run The AI Corner.
I’m not just into writing about AI. I run Allexive, and we help businesses grow without adding headcount by implementing AI platforms, and building AI systems.
👋 Mike & Erin