In partnership with
Kia ora! Welcome to New Zealand’s weekly roundup of AI news and education.
A fascinating company has emerged at the edge of the AI and robotics race.
MicroAGI is offering free apartment cleans in New York, with one catch: cleaners wear head-mounted cameras to train household robots.
It’s a reminder that AI isn’t just moving into documents, code, and search. It’s moving into the physical world. And the irony is that people thought hands-on work was safer from AI. Now footage of that work may be one of the most valuable datasets.
The service isn’t the product here. The cleaning is the acquisition cost, and the data is the asset. Article here.
Reminder: We’re hiring a Lead AI Engineer at Allexive. This is for a gun engineer who lives and breathes AI agents, and wants to help shape the technical direction of a fast-growing AI engineering business.
You’ll work across clients like Mitre 10, Petbarn, Target Furniture, RealNZ and Opes Mortgages, rolling out Claude, Copilot and ChatGPT, and building with Claude Agent SDK, Microsoft Agent Framework and Google ADK. DM me or apply here.
Happy reading ✌️
Did someone forward you this? Sign up!
🇳🇿 New Zealand News
NZ Herald reported on the global enterprise backlash against soaring AI subscription and compute costs, with companies that moved fast on AI adoption now reassessing return on investment as bills mount. Interest.co.nz separately published analysis questioning whether AI can realistically deliver the fiscal savings the government is projecting from replacing public servants with AI tools.
3 min read
Our take: The government's assumption that replacing ~9,000 public servants with AI will deliver $2.4 billion in savings carries the same flaw at scale. AI does not reduce headcount by doing what people do, faster. It reduces headcount by making certain tasks unnecessary or by enabling fewer people to cover the same scope, which requires a different kind of workflow redesign that takes time, capability, and implementation resources the savings projection doesn't account for.
The Reserve Bank of New Zealand's May 2026 Financial Stability Report named Anthropic's Mythos as an emerging risk to the financial system, the first time a central bank has identified a specific frontier AI model in an official stability document. The RBNZ warned that concentration among AI providers and the capabilities of frontier models could amplify cyber threats and harm borrowers' ability to repay loans. The Reserve Bank of Australia is conducting parallel monitoring; RBNZ is engaging with Trans-Tasman counterparts.
4 min read
Our take: Naming a specific model in a stability document is a different category of signal from "AI poses systemic risk in general”. It means the RBNZ's analysis got specific enough to identify a particular capability threshold as the concern, which is either very reassuring (they understand what they're watching) or very unsettling (they've found something concrete enough to name). For NZ business leaders, the takeaway is not "avoid Mythos". It's that central banks are now tracking AI model capabilities as a systemic variable, which means AI infrastructure decisions are no longer just a technology team question. They're a risk management and governance question.
Senior NZ government chief digital officers flew to Microsoft's US headquarters this week with AI and the future of public sector work on the agenda, deepening a relationship that BusinessDesk's investigation tracked at $20.2m in AI spending across 17 agencies since 2023. NZ Police is the largest single spender at $8.6m. Microsoft Copilot is the dominant tool across government, adopted as an add-on to existing Microsoft contracts rather than through competitive tender.
4 min read
Our take: The $20.2m figure is almost certainly an undercount. Thirteen of seventeen agencies responded, the methodology was self-reported, and several agencies are using AI embedded in broader Microsoft contracts that may not be separately categorised as AI spend. The real number, across the full public sector, is likely a multiple of what was reported. The question for NZ business leaders watching this is whether their own AI adoption is following the same pattern: Microsoft already there, Copilot added to the licence, no structured evaluation of alternatives. "We're already a Microsoft shop" is a vendor relationship, not an AI strategy.
Ivo, founded in Auckland and now headquartered in San Francisco, has become the Official Legal AI Partner of New Zealand Football, signing the All Whites and the Football Ferns as the company accelerates its global expansion. Ivo's contract intelligence platform is used by Meta, Uber, IBM, and Shopify to automate contract workflows and extract business intelligence from contract portfolios.
2 min read
Our take: The NZ Football partnership is a brand play, not a product play. A company already inside Meta, Uber, and Shopify's contract infrastructure does not need the All Whites to prove technical credibility. What NZ Football gives Ivo is a visible, internationally recognised NZ connection as the company grows, useful for recruiting, for local enterprise sales, and for the kind of founder story that travels well in the US. Ivo is a useful case study in what NZ-origin AI success actually looks like in 2026: founded here, scaled from San Francisco, global enterprise clients.
A UN University report projects AI will consume 3% of global electricity and 3.7 trillion litres of water annually by 2030, with the land footprint of that electricity use estimated at nearly 6,000 square kilometres. The Science Media Centre published NZ expert reactions the same day, with researchers warning that NZ's debate about AI infrastructure has been narrowly focused on grid supply rather than water, land, and what communities are actually trading for data centres.
4 min read
Our take: RNZ separately published an academic counterpoint arguing AI's own advances may solve the problems its adoption creates, better efficiency, better cooling, better energy management. That argument is worth tracking but it's a projection, not a receipt. The resource numbers in the UN report are based on current trajectory. "We'll fix it with future AI" is not a resource management plan.
Talk to your AI tools the way you'd talk to a colleague.
You don't send a colleague a three-word brief. You explain the context, the constraints, what you've already tried. But typing all that into ChatGPT takes forever — so you don't.
Wispr Flow lets you speak your prompts instead. Talk through your thinking naturally and get clean, paste-ready text. No filler words. No cleanup. Just detailed prompts that actually get you useful answers on the first try.
Millions of users worldwide. 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 AI is electricity, but most companies are still running the 1890 factory: Factories swapped the steam engine for an electric motor and kept everything else the same. Output barely moved for 20 years. A new generation redesigned the floor around electricity's possibilities, and productivity surged. Three examples with AI today.
7-min read.
2️⃣ Three questions to decipher whether you’re building a product that the frontier AI labs will consume: On one engagement, Claude shipped what was being built six weeks into delivery. The filter: will a stronger model alone improve this? Or does the value sit in the workflow logic, approvals, and trust built around it? Joe Schmidt IV at a16z called it the Yellow Brick Road problem.
2-min read.
3️⃣ Training teaches people to use AI. Redesigning work changes what they do: One manufacturer needed AI to pull a parts list from an engineering drawing. A tourism operator needed to connect content to revenue. Both had the tools, had been trained, and neither changed how work got done. The lighthouse team model fixes that.
2-min read.
4️⃣ Seven tests that separate a real Claude skill from a dressed-up prompt: A real skill reads actual data before running, stops cleanly when context is missing, and bakes every correction into a learnings file. Most Claude skills fail at least three. Run the first test on the skill that fires most often.
2-min read.
🎙️ The AI Corner Podcast
This week's guest is Loren Thomas, Head of People Strategy and Experience at Trade Me. Hear:
Why Loren wishes she'd scaled foundational AI training faster, and what the spike in adoption data actually showed.
How Trade Me's "blast radius" framework separates personal AI tinkering from high-stakes business-owned projects.
Why AI transformation isn't a tech initiative — and why people leaders need to take the wheel.
🌍 Tech Updates From Global
The selected top headlines from each major AI tech company.
Anthropic
Anthropic filed a confidential IPO prospectus with the SEC, targeting a potential Wall Street debut as soon as autumn 2026. (Jun 1)
Anthropic called on major AI labs to consider a verifiable, coordinated development pause, warning AI is advancing faster than human oversight. (Jun 5)
Anthropic launched its Claude Partner Network Services Track and Partner Hub, backed by $100M investment, with 10,000+ consultants certified since March. (Jun 3)
Claude Mythos Preview expanded to 150+ organisations across power, water, healthcare, communications, and hardware sectors in 15+ countries, adding Claude Security for codebase scanning. (Jun 2)
Claude Agent SDK and
claude -pusage will draw from a separate monthly credit pool from June 15, priced at standard API list rates. (Jun 2)
OpenAI
OpenAI expanded Codex beyond developers with six role-specific plugins across Snowflake, Figma, Salesforce, and 59 more apps, plus a "Sites" feature for deploying web apps via URL. (Jun 3)
ChatGPT memory doubled in capacity for Plus and Pro users, adding "dreaming" (passive memory revision over time) and a reviewable memory summary page. (date unconfirmed)
ChatGPT Lockdown Mode launched for all logged-in users, disabling browsing, deep research, and agent mode to reduce prompt injection risk. (date unconfirmed)
Microsoft
Microsoft Scout launched at Build 2026 as an always-on personal AI agent running across Teams, Outlook, OneDrive, and SharePoint. (Jun 2)
Microsoft Autopilots unveiled, a new class of always-on autonomous agents with their own identity that act independently across Microsoft 365 and Azure. (Jun 2)
Microsoft IQ, a knowledge and context layer feeding agents real workplace data and web grounding, reached general availability across Copilot, Foundry, and Copilot Studio. (Jun 2)
Azure AI Foundry Hosted Agents announced for general availability by end of June, with four new production-ready capabilities added. (Jun 2)
OpenAI and Microsoft confirmed compliance with a Trump executive order requiring 30-day advance government access to new AI models before public release. (Jun 5)
Gemini 3.5 Pro nearing general availability with a 2-million-token context window and a "Deep Think" reasoning mode first shown at Google I/O. (Jun 6)
Gemma 4 12B released, the first mid-sized Gemma model with native audio input, designed to run locally on 16GB RAM consumer laptops. (date unconfirmed)
Google Workspace added Ask Gemini in Drive with Gmail sources, Organise My Files, and raised Studio usage limits for AI Expanded Access licence holders. (Jun 5)
NVIDIA
NVIDIA RTX Spark unveiled at Computex, its first Arm-based Windows laptop chip targeting the $200B PC CPU market, with OEM devices from ASUS, Dell, HP, Lenovo, and MSI arriving this northern autumn. (Jun 1)
xAI
Grok Voice and Grok Imagine 1.5 Preview launched, with Imagine 1.5 reaching number one on the Artificial Analysis Image-to-Video leaderboard and video extended to 15 seconds. (Jun 4)
Grok Skills went live across web, iOS, and Android, letting users build reusable workflows and output formats on Grok 4.3. (date unconfirmed)
xAI paused hiring for Grok training specialists, signalling a shift in its fine-tuning approach. (Jun 3)
Amazon
OpenAI GPT-5.5, GPT-5.4, and Codex reached general availability on Amazon Bedrock for enterprise AI workloads. (Jun 3)
Amazon Bedrock launched a redesigned developer console supporting the OpenAI Responses API, OpenAI Chat Completions API, and Anthropic Messages API simultaneously via a unified endpoint. (date unconfirmed)
Amazon Connect expanded into four agentic AI solution sets covering supply chain, talent acquisition, customer experience, and healthcare workflows. (date unconfirmed)
Perplexity
Perplexity Computer expanded to Windows devices after launching on Mac. (date unconfirmed)
Canva integrated into Perplexity Computer, letting users move from AI research to ready-to-publish creative assets in a single workflow. (date unconfirmed)
Perplexity CEO told CNBC the company has tripled annualised revenue since January 2026, naming "token value per watt per user" as the defining AI race metric. (Jun 3)
Meta
WhatsApp Business AI agent launched globally after two years of testing in India and Mexico, handling support, product recommendations, appointment booking, and sales lead qualification. (Jun 3)
✨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