Kia ora! Welcome to New Zealand’s weekly roundup of AI news and education.
Check out the super relatable video below. It captures a shift we’re all feeling: we’re no longer just asking, “is this AI?”. We’re asking, “can I trust this?”.
Generated? Edited? Scripted? Enhanced? Or completely real, but suddenly suspicious because our trust has shifted?
That’s the messy middle we’re heading into. Because when trust drops, people don’t just doubt content. They begin to doubt the person, brand, or institution behind it.
Most leaders think hackathons are ideation events.
Here is a rundown of the preparation process I went through for our hackathon on Friday, and the exact framework we use to turn a standard event into an operating-model upgrade.
Happy reading ✌️
Did someone forward you this? Sign up!
🇳🇿 New Zealand News
Aimer Farming, founded by ex-DairyNZ scientist Dr Jeremy Bryant, secured NZ$600,000 of a NZ$1.675m project from the Ministry for Primary Industries to expand its AI pasture measurement software across hundreds of NZ dairy and beef farms. Already on 650+ farms with 10,000+ pasture measurements per week. Next phase introduces "Ask AIMER", a chat-based assistant inside the app, integrating with virtual fencing systems for precision grazing.
3 min read
Our take: The "Ask AIMER" feature, which is a chat interface running against Aimer's accumulated data, is the playbook every New Zealand vertical-software company should be following this year, because the competitive advantage is not the underlying AI model given that anyone can rent GPT-5 or Claude on similar terms. The actual advantage is the proprietary dataset, and Aimer has been collecting 10,000 pasture measurements per week from real farms for years, which makes its data extremely difficult to replicate, and New Zealand's edge in vertical AI is going to come from datasets like this rather than from competing on models.
Infratil's CDC subsidiary signed a 30-year, 555MW contract with a US customer to build Australia's largest data centre, the company's largest-ever contracting announcement. It landed days after Amazon wrote down its abandoned NZ data centre site. Australia got the AI compute build-out, NZ got the write-down.
3 min read
Our take: On the same day this contract was announced, RNZ ran a feature asking who actually benefits when Datagrid builds a 280MW AI factory near Invercargill, and the honest answer is that not many New Zealanders do, because even when New Zealand wins a build the workloads tend to serve global customers and the local gain is mostly construction jobs plus a long-running power bill. The CDC contract heading to Australia is just the same trend in clearer view, with hyperscaler customers caring about latency to their user base rather than which country has the cleaner electricity.
AI Forum NZ Executive Director Madeline Newman launched the refreshed AI Blueprint for Aotearoa with a 2030 vision and a review of 28 prior reports finding adoption between 40% and 80% across sectors but persistent low trust. Six priority sectors: agriculture, AEC, creative, education, environment, health. Two new focus areas: Social Licence and Sustainable AI. National conversations run April-June ahead of the September Aotearoa AI Summit.
5 min read
Our take: While I support it, a target of global AI leadership by 2030 is ambitious when competing in a category dominated by the US and China. The most useful and important parts of the Blueprint are the explicit list of six priority sectors covering agriculture, construction, creative industries, education, environment, and health. Those are the six places where New Zealand has real domain knowledge and real datasets, and if the next twelve months of funding concentrate there rather than spreading thinly, the Blueprint becomes something businesses can act on.
A Massey University pilot study (10 academics, interviews across film, music, design and digital art) presented at the AI & Creativity Summit Wellington on 6 May found GenAI is absorbing the low-paid, freelance and unpaid work juniors traditionally use to build portfolios. No interviewee could cite a specific paid-role displacement to date. Lead researcher Associate Professor Dave Carter and researcher Gwen Isaac warn the sector against being a "passive receiver of big tech's plans".
3 min read
Our take: The Massey researchers went into this study expecting to find significant job losses across New Zealand creative industries and came back with no specific layoffs to point to, because none of the interviewees could name a colleague who had been let go because of AI, although every one of them could describe the kinds of work that have already disappeared. AI will absorb that bottom rung first across every knowledge profession before it touches the senior roles, which means New Zealand businesses hiring graduates need to plan now for how they train the next generation when most of the traditional entry-level work has been pulled out of the pipeline.
📚️ Mike’s Takes From The Week
Helping leaders and teams adapt, learn, and scale with AI.
1️⃣ AI Cheatsheet: Ran an AI enablement session and walked out reminded that most people only use 5% of what they pay for in AI tools. Here is a categorised cheat sheet of 30+ prompts that unlock features hiding inside ChatGPT, Claude, Gemini, and Perplexity.
3-min read
2️⃣ Experimentation is the key to AI adoption, but is often guided poorly by Execs: The CEO who rolled out Copilot six months ago is now being asked by the board what changed. Why forced adoption produces compliance theatre, and the "let them gravitate, then constrain" methodology of experimentation that drives real ROI.
3-min read
3️⃣ Building an AI-native company: Y Combinator just put out a piece on how to build an AI-native company, and most leaders will read it wrong. The harder question is what AI-native looks like inside a 5,000-person business that cannot restart from scratch. Sharing my write up on how to become more AI-native.
2-min read
4️⃣ Elon vs Sam, week 1: Possibly the most important AI trial of the decade. $134 billion in damages, a courtroom gasp, and a diary entry that turned the lawsuit into a trial. Almost none of it is actually about AI, it's about who owns the next ten years.
8-min read
5️⃣ The AI bubble story has been quiet for months, and Q1 2026 explains why: $1.45 trillion in audited backlog. Three CEOs saying supply can't catch demand. Six structural reasons this isn't a bubble, plus the two real risks that did survive Q1.
7-min read
🎙️ The AI Corner Podcast
This week's guest is Justin Flitter, returning guest and one of the voices helping shape AI adoption in New Zealand. Hear:
Why most businesses now have access to AI, but results are still scattered and inconsistent.
How SMBs and mid-market companies can move from experimentation into trusted, everyday use.
Why literacy, confidence, and small hands-on proof points matter more than another generic training session.
🛠 Latest Builds and Finds
Helping advanced builders stay at the frontier of AI.
1️⃣ Garry Tan's "fat skills, thin harness" thesis. The YC CEO has open-sourced GBrain, a 100,000-page personal knowledge system running ~100 crons a day. The model is interchangeable, the harness is dumb on purpose, and the compounding asset is the fat data plus a library of skills built by a meta-skill called Skillify. This is exactly the architecture I've been circling with my own vault.
Article
2️⃣ Peter Steinberger’s Peekaboo 3.0 lets Claude Code drive native Mac apps. His MacOS automation toolkit pairs pixel capture with the accessibility tree and ships an MCP server, so the model gets semantic targets instead of guessing at coordinates. MIT, 3.5k stars, works with every major provider and local Ollama. Evaluating it for the Allexive bits where native apps don't expose APIs.
Peekaboo.sh
3️⃣ Thariq on why HTML, not markdown, should be the default agent output format, which we use for all presentations. His point is that markdown caps out past 100 lines, where HTML carries tables, SVG, interactions, and is one S3 upload away from a shareable link. We’ve shifted all of Allexive presentations and interactive discussions to a HTML experience due to the vastly improved interactive it offers without limitation.
Article
🌍 Tech Updates From Global
The selected top headlines from each major AI tech company.
OpenAI
The Deployment Company, a $10B Delaware-domiciled JV with TPG, Brookfield, Advent, Bain and 15 other PE backers, finalised on May 4 with $4B+ raised, OpenAI retaining super-voting control and a guaranteed 17.5% annual return over 5 years for the consortium. (May 4)
ChatGPT for Excel and Google Sheets graduated to GA on GPT-5.5 across Free, Go, Plus, Pro, Business, Enterprise, Edu and K-12 plans, with Business and Enterprise on free preview through June 2 before plan-credit metering. (May 5)
GPT-5.5-Cyber rolled out in limited preview to vetted Trusted Access for Cyber members for vulnerability triage, patch validation, malware analysis and reverse engineering, with credential-theft and malware-writing tasks still blocked. (May 7)
ChatGPT memory added a sources view (across all models) and pulls more aggressively from past chats, saved memories, files and Gmail for Plus/Pro, closing a long-running personalisation transparency gap. (May 5 to May 7)
Anthropic
Code w/ Claude in San Francisco shipped three Managed Agents capabilities, multi-agent orchestration, outcomes (success criteria for self-iteration) and "dreaming" (between-session memory consolidation), with Harvey reporting ~6x task-completion lift from dreaming and API volume up 17x year-over-year. (May 6)
SpaceX (now SpaceXAI) gave Anthropic full Colossus 1 compute capacity in Memphis, 220,000 Nvidia GPUs and 300MW within the month, with both sides signalling interest in multi-gigawatt orbital data centres after Musk earlier called Anthropic "evil". (May 6)
Anthropic, Blackstone, Hellman & Friedman and Goldman Sachs formed a $1.5B AI-native enterprise services JV with $300M each from Anthropic, Blackstone and H&F to embed Claude and Anthropic engineers in mid-market PE portfolio operations, directly targeting consulting. (May 4)
Anthropic launched 10 ready-to-run financial services agent templates (pitch decks, financial-statement review, compliance escalation) with Microsoft 365 add-ins for Excel, PowerPoint, Word and Outlook coming and a Moody's MCP app for banks, insurers and asset managers. (May 5)
Google / DeepMind
Gemini in Google Docs added persistent custom instructions for tone, style and formatting (capped at 1,000 active instructions per account) for AI Plus and above, US English-only over a 15-day rollout. (May 5)
Macquarie Bank reported 130,000 hours saved over 7 months using Gemini Enterprise (per Gemini Enterprise release notes), as new direct-file-generation features added Docs, Sheets, Slides, PDF, DOCX, XLSX, CSV and LaTeX outputs. (May 8)
Microsoft
Copilot Cowork launched on iOS and Android with a plugin ecosystem (LSEG, Miro, monday.com, S&P Global, Fabric, Dynamics 365) for Frontier customers to delegate multi-step work from mobile. (May 5)
Microsoft is internally weighing whether to delay or abandon its 2030 100% hourly clean-energy matching pledge as AI data centre buildout outpaces clean-power supply (per Bloomberg). (May 6)
xAI / SpaceXAI
Musk announced xAI will be dissolved as a separate company with Grok and future AI products shipping under "SpaceXAI", finalising the structural fold-in after SpaceX's earlier $250B acquisition. (May 7)
SpaceXAI gave Anthropic full Colossus 1 compute capacity (220,000 GPUs / 300MW) at its Memphis facility, with both companies flagging interest in multi-gigawatt orbital data centres. (May 6)
Amazon / AWS
Amazon Bedrock AgentCore Payments launched in preview with Coinbase and Stripe, letting AI agents transact autonomously via USDC stablecoin or fiat using Coinbase's x402 protocol and Stripe's Privy wallet, starting with API and paywalled-content micropayments. (May 7)
Andy Jassy used a CNBC post-earnings interview to defend the $200B+ 2026 capex plan, framing AI as "once-in-a-lifetime" and pointing to AWS's $15B AI run-rate growing ~260x faster than cloud's early days. (May 4)
✨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