Kia ora! Welcome to New Zealand’s weekly roundup of AI news and education.
I’ve now coached two CEOs and I’m currently coaching two more with very big remits. It’s been a fascinating learning process for me. Here’s what I’m learning about how executives use AI:
Their work involves absorbing information, weighing competing views and deciding what to do. I’m finding that coaching is focused on building tools to support their thinking and judgement, with much less emphasis on producing documents or automating tasks.
The latest report needs to be read alongside earlier commitments, the reasons behind decisions and the assumptions they depended on. So the context we build has to be much more deliberate and mature in approach: much more specific to their responsibilities, maintained over time and easy to retrieve.
Form a view first, then ask AI to assess the original material before sharing that view. Comparing the two can reveal missing evidence, an overlooked option or an assumption worth testing. Leveraging existing tried and true frameworks to manage bias is a common theme coming through.
Alongside the standard four-week 1:1 AI Coaching Programme I run with folks, I’m running a new specialised 1:1 AI Executive Coaching Programme for C-suite leaders in larger companies. We work on real decisions (NDAs signed), meeting preparation and the information you rely on to lead. It’s personal, practical coaching designed to help you examine options, challenge assumptions and recover time for thinking. Reply for the details.
I’m also designing a programme for Executive Assistants. If you’re an EA already using AI at work, I’d love to chat.
Happy reading.
Did someone forward this? Sign up.
🇳🇿 New Zealand News
The Greens propose registers of government AI and new rules for very large data centres. The election policy includes human oversight, Māori data protections and a consent pause until regulation exists.
Read
Our take: AI policy needs to distinguish demonstrated harms from predictions of what might happen. I’m concerned alarming headlines are encouraging restrictions before NZ gives the technology a serious chance. The Greens’ plan reads as prematurely restrictive. Keeping data onshore while pausing hyperscale data centres needs explaining. Children need AI skills alongside foundational learning, and local models need a credible investment case. I support proportionate oversight, but where is the equally ambitious commitment, from any party, to help NZ use this technology well?
Xero says 94% of surveyed NZ accounting practices expect no AI-driven team cuts. Its survey of 526 senior professionals finds 49% use AI in everyday work and 61% say it has changed hiring.
Read
Our take: The bigger opportunity is retaining and reusing what a business learns. Capture how teams work, make decisions and serve customers, then use that shared context to support judgement across the company. Over time, AI can help identify patterns and opportunities for new services, products and markets. That accumulated knowledge can become a proprietary asset, with value beyond individual productivity.
The Reserve Bank and FMA have set expectations for AI-related cyber risks in finance. The Financial Markets Authority and Reserve Bank want firms to address oversight, supplier dependence and security controls under existing obligations.
Read
Our take: The Reserve Bank is right to focus on the basics. Who removes access when someone leaves, checks it happened and records it? Who owns software accounts, costs and cancellations? Shared logins and forgotten permissions are existing business problems. Connected AI can inherit that access. For firms with these gaps, modernisation needs to come before giving AI broader authority.
SEEK reports AI skills in 4.2% of NZ job advertisements in September. Mentions rose 88.2% over a year, while ads for roles assessed as easier to automate fell 2.6%.
Read
Our take: “AI skills” is close to meaningless without a level attached. Chatbot questions, email rewrites and summaries are barely a starting point now. The next step is delegating defined work to AI and checking the result; it doesn’t require building an agent. Recruiters and job platforms need to explain and assess that progression. The gap between what AI can do and how people use it is still enormous.
The NZ Defence Force expects an Arcadia agreement with Five Eyes partners early next year. The shared military data and AI project still needs an approved budget, with NZ funding expected to be sought in Budget 2027.
Read
Our take: I don’t support war, but I do support equipping NZ to defend itself and working with allies on AI development, even as a means for learning. National security deserves deliberate investment in local capability. Kiwi entrepreneur Sean Gourley’s work through Primer shows the calibre of talent we can contribute. I’d like to see NZ develop more of that expertise here, alongside access to allied technology.
AI Forum has announced Te Kāhui Māori, a platform for Māori voices in AI. The website shares achievements and perspectives following discussions about Māori leadership and data sovereignty at the Aotearoa AI Summit.
Read
Our take: A public profile can help an organisation find Māori expertise before it chooses a supplier, and gives key Māori voices a platform to continue to educate the public and Government on the risks of ignoring how AI can negatively impact Māori.
National is retaining its AI strategy while Labour says its action plan would replace it. RNZ reports differing views on oversight and questions about MBIE using AI for ideas when preparing the existing strategy.
Read
Our take: Additional oversight is needed, but it needs to come with a much stronger commitment to developing the industry. I want National and Labour to explain what investment and resources they will provide for research, talent and businesses putting AI to work. Regulating the technology is only part of the job. NZ also needs the capability to build with it and benefit from it.
NZ technology users group TUANZ is calling for a cross-party national digital strategy. Chief executive Craig Young argues election-cycle changes undermine investment, while the group republishes proposals on access, security, data and workforce development.
Read
Our take: A cross-party document needs commitments that survive changes in ministers and budgets. Published funding, delivery dates and named agencies would give businesses a stronger basis for investment.
AUT says more than 4,000 staff can access models through AUTGPT. Its institutional recap of a 30 September meetup describes Claude, ChatGPT and Gemini access, without reporting measured productivity gains.
Read
Our take: Staff access becomes useful when a team changes a recurring task together, and those shared learnings are propagated across the group. A course administrator could test drafting student replies while colleagues check accuracy and record the review time.
BERL examines who could benefit from AI-assisted breast screening. Navdeep Kaur’s analysis considers how possible efficiency gains could improve access, without announcing a new NZ deployment or measured local result.
Read
Our take: Faster image reading helps patients only when the next appointment or treatment can happen sooner. Screening teams should measure the complete journey before claiming shorter waits.
📚️ Mike's Takes
Helping leaders and teams adapt, learn, and scale with AI.
1. Shared AI workflows need maintained instructions and named owners: Mike identifies duplicated skills, conflicting templates and corrections that remain in private chats. Shared procedures need owners who approve changes, while teams measure whether completed work saves time or money and keeps improving for colleagues.
Read
2. Meta Muse, Grok Bot, ChatGPT Dots: Businesses need to decide who operates their employees’ personal AI agents: An employee’s assistant could prepare proposals while retaining company information and acting through connected apps. Mike examines company-operated assistants, shared instructions and specialist agents, with permissions enforced by business systems and approved corrections available to colleagues.
Read
3. Business and tech teams disagree on how AI should be rolled out: Business teams want useful tools, while IT must check access, costs and what happens when systems fail. Mike proposes agreeing which work to improve, what can start now, what needs resolving and who owns the next decision.
Read
4. Leaders need direct experience using AI on their own work: Preparing systems and controls still leaves leaders needing practical experience. Mike describes CEOs seeking coaching to understand what Claude and ChatGPT can do, then argues that leaders should test real work before deciding how the business will change.
Read
The AI Corner Podcast
This week's guest is Andrew Wierda, CEO and co-founder at Nylon. Hear:
Why tax research requires a wider understanding of law: Andrew explains how questions about employees, trusts and cryptocurrency pushed Nylon beyond its original tax focus.
How Nylon connects specialist research to Claude through MCP, while mapping overturned cases in advance and keeping each client’s documents and emails within separate engagements.
Hear Andrew’s hiring principle: recruit people who enjoy solving problems and can learn quickly, because attachment to today’s methods becomes a constraint as their work changes.
🌍 Tech Updates From Global
Recent releases from the companies building AI.
Anthropic
Code Review becomes generally available for Team and Enterprise, with automatic GitHub reviews and repository-specific Quick, Standard or Deep settings. (Oct 15) Read
Eligible Enterprise organisations can claim at least US$10,000 in Code Review credits, which expire 30 days after claiming. (Oct 15) Read
Claude Platform adds OpenTelemetry tracing in public beta across Managed Agents, the Messages API and SDKs for monitoring agent actions. (Oct 14) Read
Claude Code adds idle compaction for large inactive conversations, reducing context processing when work resumes after the prompt cache expires. (Oct 14) Read
Claude Code adds controls for idle compaction and the client safety classifier model, giving administrators more control over unattended sessions. (Oct 16) Read
Claude Code enables forked subagents by default across its SDK, desktop app and VS Code, extending parallel task execution. (Oct 16) Read
Anthropic opens its fifth Asia-Pacific office in Singapore and signs a government agreement covering AI safety, evaluation and responsible adoption. (Oct 14) Read
Anthropic gives six METR evaluators employee-like access, with publication rights and defined restrictions on confidential information. (Oct 15) Read
Anthropic demonstrates replication steps in controlled experiments on consumer-device simulations, while reporting that no complete rogue replicator has been demonstrated. (Oct 12) Read
OpenAI
ChatGPT Slides starts rolling out to eligible Pro, Business and Enterprise users, with shared editing and presentation exports. (Oct 16) Read
GPT-6 Pro adds faster responses, improved tool use and parallel agents for eligible ChatGPT Pro users. (Oct 16) Read
Prompt-to-voice becomes available to all API customers, creating reusable custom voices from text descriptions at standard speech-generation rates. (Oct 16) Read
GPT-Live adds native Decisions integration, allowing applications to act during speech or respond to incoming and outgoing transcript updates. (Oct 15) Read
Codex CLI adds searchable settings, allowing supported preferences to change immediately and identifying changes that require restarting. (Oct 15) Read
Codex CLI allows immediate steering during responses, with interruptions also stopping active code execution and nested tool calls. (Oct 15) Read
Codex CLI enables eligible Daybreak security access by default and adds parallel execution for independent read-only tools. (Oct 15) Read
Iowa joins seven other US states suing OpenAI over alleged consumer deception and chatbot harms, with allegations awaiting court determination. (Oct 14) Read
OpenAI publishes a Third Moment Research case study describing 10,000 concurrent agents working on investigations lasting 36 to 48 hours. (Oct 14) Read
Gemini 4 Argon becomes generally available through the Gemini API and Google Cloud for complex coding, scientific and business tasks. (Oct 13) Read
Antigravity coding agents join the Gemini macOS app, building and testing projects in the background after a plan is approved. (Oct 13) Read
Gemini adds a Customise hub for apps, skills and response preferences, with most connected services initially available in US English. (Oct 13) Read
Skills begin rolling out in the Gemini app, while Google delays skills inside Workspace until a revised launch schedule is confirmed. (Oct 13) Read
Google Meet announces post-meeting routines for drafting follow-up emails or adding action items to Google Tasks when note-taking is enabled. (Oct 12) Read
Gmail AI Inbox begins expanding to paid Workspace customers in US English, extracting suggested tasks and summarising priority topics. (Oct 15) Read
Gmail Live starts reaching eligible Workspace customers on iOS and Android in US English, with cited email answers and selectable voices. (Oct 15) Read
Universal Cart launches in US Search and Gemini, collecting products from multiple merchants and tracking prices before checkout. (Oct 15) Read
Google demonstrates Gemini 4 Argon controlling robots and constructing simulations from images, publishing examples rather than announcing a separate robotics model. (Oct 13) Read
Chinese Frontier Labs
DeepSeek upgrades V4.1 Flash across its app, website and API, reporting stronger reasoning and coding performance at unchanged API prices. (Oct 17) Read
Open-Weight Model Updates
Meta releases Muse Spark Open with 816 billion parameters and a million-token context window, under a licence with limited commercial terms. (Oct 15) Read
Muse Spark Open supports local deployment across two Apple M5 Ultra Mac Studios using MLX, alongside hosted inference providers. (Oct 15) Read
NVIDIA releases Nemotron 3.5 Super VL, a downloadable multimodal model for document analysis, visual interfaces and extended agent tasks. (Oct 15) Read
Microsoft
Copilot begins rolling out legal plugins from providers including LawVu, Harvey and Docusign through its centrally managed plugin registry. (Oct 15) Read
GitHub AI Scan becomes generally available for pull requests, adding advisory vulnerability findings beyond CodeQL’s supported languages and frameworks. (Oct 15) Read
Gemini 4 Argon becomes generally available in GitHub Copilot, with gradual rollout across supported editors, its app and coding agents. (Oct 13) Read
Copilot administrators can require fresh managed settings at startup, preventing supported clients from running when current policies cannot be confirmed. (Oct 13) Read
GitHub in Slack and Teams adds multi-repository conversations, supporting issue creation, code changes and bug investigation across repositories. (Oct 15) Read
Meta
Meta launches Muse Wonder through its model API and Muse Code, with Wonder Full available to Muse Code subscribers. (Oct 12) Read
xAI
Grok Voice 3.0 becomes available through the speech-to-speech API at unchanged pricing, with the latest-model alias updating automatically. (Oct 14) Read
Reports & Data This Week
BCG finds four in ten employees want greater CEO transparency about AI’s effect on careers, surveying 500 employees and 645 CEOs. (Oct 14) Read
👋 Mike & Erin