Kia ora! Welcome to New Zealand’s weekly roundup of AI news and education.
A CEO I’m working with wanted some of his Sunday back. We built a /1:1-prep Skill to find each direct report’s agenda, research the talking points and bring forward unfinished commitments.
Then we gave each agenda item its own research agent.
Each calls his /company-foundations Skill to check the relevant strategy, budget and leadership objectives before assessing the latest emails, SharePoint papers and Teams updates.
A coordinating agent brings the findings into one briefing.
We started with a Sunday schedule.
Then he pointed out that Wednesday’s agenda might arrive on Tuesday night. So we changed it to 5am daily, refreshing each briefing as new information arrives and to-dos are completed.
That’s two of the five Skills we’ve been developing. I’ve written up five key Skills every Executive should be building and developing for their own work.
🎧 JD Trask returns to The AI Corner podcast this week. After setting the Aotearoa AI Summit 2026 alight earlier this year, we talked about New Zealand’s lack of ambition on AI, where he thinks politicians are getting regulation wrong, and why employers who keep delaying AI could lose their best people.
Want help moving from using AI as a chat bot to delegating tasks? I’m now running:
A four-week 1:1 AI Coaching Programme works around your tools and actual workflows, and
1:1 AI Executive Coaching Programme is open for C-suite leaders in larger companies.
Reply for the details.
Happy reading ✌️
Did someone forward this? Sign up.
🇳🇿 New Zealand News
Chlöe Swarbrick puts the odds of AI catastrophe at 50/50. In Finn Hogan’s Stuff interview, the Greens co-leader describes a pause on new hyperscale data centre consents until safeguards exist.
Read
Our take: Chlöe puts AI catastrophe at 50/50 while discussing a policy programme that includes pausing new hyperscale data centre consents. Discussing them together encourages readers to see the restrictions as a response to the danger, even though the stated conditions for data centres concern environmental safeguards. That deserves much harder scrutiny than pointing to prominent researchers who are worried too.
Geoffrey Hinton said in 2016 that people should stop training radiologists. A decade later, the Royal College of Radiologists is calling for more training to address shortages. Being brilliant at developing AI does not make someone a reliable forecaster of what happens when it enters the economy. That failed advice does not disprove catastrophe risk, but it should end the habit of treating a famous researcher’s confidence as sufficient evidence.
JD Trask makes the economic objection in this week’s podcast: push the infrastructure offshore and New Zealand risks becoming a customer paying everyone else. Put evidence behind the 50/50 claim and explain the expected benefits and costs of the proposed restrictions.
 |  |
AUT student Hunter Yates, who works with us at Allexive, questions what a degree proves when AI can complete assignments and tutor students. Alex Richter agrees assessment needs to change, but argues university should also develop understanding, relationships and judgement, beyond the qualification itself.
Auckland Council received more than 13,000 reports through its AI reporting tool. Ask Auckland Council, launched in April, sorts residents' descriptions and photos before directing reports to council teams.
Read
Our take: This is the sort of AI use councils should be getting on with: let people describe a problem without understanding the council’s internal departments. Council reports that the service saves users around nine minutes compared with calling, which is a concrete reason to keep improving it. Now publish whether those reports reach the right team and get resolved faster, because a quicker complaint followed by the same delay is only part of the job done.
RNZ reports that job applicants are often left without a response. Employment Hero says 44% of its surveyed workforce experienced ghosting, while its NZ manager links some employer pressure to AI-generated applications.
Read
Our take: Blaming AI for candidates being ghosted is a bloody convenient explanation for employers who still owe people a response. Higher application volumes can make recruitment harder, but the technology to acknowledge applications and send rejection emails existed well before ChatGPT launched. If AI makes polished applications less useful, change the assessment and ask candidates to demonstrate the work. Leaving people waiting tells them more about the employer than the application.
NZ publishers want Meta to withdraw proposed content-crawling terms. NPA seeks permission and payment for AI use of journalism, while European publishers say Meta postponed the terms to 20 November.
Read
Our take: Meta should pay for journalism it uses to build commercial AI products, and publishers should be able to promote their reporting without accepting every additional use Meta proposes. Reporting costs money before a platform gets anywhere near it: someone attends the meeting, checks the claim and asks the awkward follow-up. If AI companies want that work, negotiate a licence rather than treating access as permission to use it however they like.
A NZ board opinion piece asks directors to own decisions about AI use. The guest article discusses suppliers, data handling, staff changes and who is responsible when an AI system fails.
Read
Our take: Too many boards lack people who have actually built technology businesses or led a digital transformation, which leaves them dependent on management and suppliers to explain the decisions they are supposed to challenge. Another AI briefing will not replace that experience. Appoint people who can question the architecture, costs, delivery plan and commercial assumptions because they have been responsible for those decisions themselves.
Other NZ stories
Karaitiana Taiuru calls for Māori-led AI education and governance pathways. His commentary responds to a rangatahi poll in which 58% said government preparation for AI was insufficient. Read
Infratil's Boyes is calling for national rules on NZ data centres. B2B News reports his call for consistent planning requirements, which the government has not adopted. Read
📚️ Mike's Takes
Helping leaders and teams adapt, learn, and scale with AI.
Computer use lets AI do the clicking. What changes at work? AI can now click, type and move information through websites and supported desktop apps. Mike shows what computer use means for everyday work, from newsletter sign-ups to collecting research, and where instructions, access and human checks still need setting up.
Watch
Use AI to think through strategic decisions before asking for answers. Drawing on a Harvard Business Review article, Mike explores how executives and individual contributors can use AI to make better strategic decisions. Explore more options, keep information current and challenge assumptions before settling on an answer.
Watch
AI agent swarms are coming. Is the business ready? Teams of AI agents could investigate checkout failures, build campaigns and propose budget changes while people set direction and review the results. Mike explains the shared context, Skills, tools, governance and runtime a business needs to make agent swarms work.
Read
Why are so few people using AI Skills for whole jobs? Faster writing still leaves the copying, formatting, filing and follow-up. Mike argues that more people should use AI Skills to handle the whole job, with connected apps, saved business instructions and approval limits, then reuse what worked next time.
Read
🎙️ The AI Corner Podcast
JD Trask, co-founder and CEO of Autohive, returns to The AI Corner. Hear:
Why JD thinks New Zealand should aim to achieve more with five million people, instead of starting every AI discussion with fears about job losses.
His objection to the Left bloc AI policy proposals: human-escalation and explanation requirements that he argues could restrict adoption, alongside too little attention to customers’ access to their own data.
His warning to employers who keep delaying: curious, capable staff will move to businesses where they can experiment with AI and change how work gets done.
🌍 Tech Updates From Global
All verified company updates for 26 October to 1 November, with each company’s priority six first in bold.
Anthropic
Anthropic said an unreleased model assembled five exploit chains against major browsers and operating systems, after coordinated disclosure to affected vendors. (Oct 27) Read
Anthropic filed an S-1 for a proposed Nasdaq listing under ANT, with the share count and offer price still undetermined. (Oct 26) Read
Broadcom and four financial partners announced US$63 billion of financing for AI equipment to be leased to Anthropic, with deliveries expected from 2027. (Oct 26) Read
In a controlled study, 50 professional analysts followed their Claude agent’s secretly assigned preference on 86% of analytical choices. (Oct 27) Read
Claude Enterprise made smart reports, in-app surveys and pooled group budgets generally available for usage analysis and team spending control. (Oct 28) Read
Anthropic renamed Max 20x to Ultra and removed the five-hour usage limit for verified accounts, keeping the US$200 monthly web price and weekly limit. (Oct 30) Read
OpenAI
GPT-6.1 Terra launched for visual reasoning and computer use in Work, Codex and the API, with standard API prices half those of prior Terra. (Oct 26) Read
OpenAI stopped new custom GPT creation in affected Enterprise workspaces on 26 October, with Plugin Creator the replacement for new workflows. (Oct 26) Read
OpenAI reorganised its public Model Spec around five values and added guidance on context, oversight and human agency. (Oct 29) Read
OpenAI doubled Work and Codex allowances for Pro 100 and Pro 500 plans and cancelled the announced Pro 200 reduction. (Oct 29) Read
CodeRabbit committed US$1 million to free security scans, offering up to US$5,000 per organisation while joining OpenAI’s Daybreak Defense Network. (Oct 26) Read
OpenAI updated its concerns policy so employees can report safety issues directly to the OpenAI Foundation board’s Safety and Security Committee. (Oct 30) Read
Google Threat Intelligence observed attackers using multi-agent offensive frameworks during real intrusions, including lateral movement and preparation for data theft. (Oct 30) Read
Google unveiled Gemini 4 Carbon for long-running coding and cyber defence, initially limiting access to selected defenders through Fairwind. (Oct 28) Read
Expanded ads in Google AI Mode to New Zealand, Australia and eight other markets, adding sponsored placements to AI search results. (Oct 28) Read
The Gemini app now creates, previews and edits documents, spreadsheets, presentations, PDFs and code within chat, with PDF editing on mobile. (Oct 29) Read
Google and Oracle agreed to offer Google TPUs on Oracle Cloud Infrastructure from 2027, creating a second hyperscaler route for TPU compute. (Oct 27) Read
Google made Gemini Computer Use generally available across its Interactions and GenerateContent APIs, enabling production browser and interface automation. (Oct 30) Read
Alphabet reported 25% year-on-year revenue growth and 103% Google Cloud growth, citing AI infrastructure and solutions as major Cloud drivers. (Oct 28) Read
Gemini in Google Slides can now edit multiple slides or an entire presentation from one prompt, changing native editable slides. (Oct 30) Read
Meta
Meta priced US$25 billion in common shares through forward agreements and US$12.5 billion in depositary shares as it expands AI infrastructure. (Oct 27) Read
Meta updated its Andromeda ad retrieval system, reporting 100 times the 2024 throughput and a 2.4% Facebook conversion increase since 2025. (Oct 28) Read
Meta deployed AI-generated tests in production and says they caught 74 serious privacy bugs before deployment in the first half of 2026. (Oct 29) Read
Microsoft
Microsoft made its shared Copilot agent runtime available through Foundry, letting enterprises build managed agents with model choice, tools and sandbox controls. (Oct 28) Read
Microsoft announced Autopilot, a persistent cloud-based personal agent with its own account and files, with a Frontier preview planned for November. (Oct 30) Read
The Copilot desktop app previewed computer use on macOS and Windows, allowing approved agents to operate applications beyond APIs and command lines. (Oct 28) Read
xAI
Grok Bot connected to 1Password without a shared vault, requiring login approval while filling credentials without exposing them to the Bot. (Oct 26) Read
Grok Bot added shopping cards, purchase approvals, shipment tracking and Shop Pay integration, alongside templates for monitoring and cancelling subscriptions. (Oct 27) Read
Grok Bot added imports of preferences and context from other AI assistants, alongside question cards that collect several answers. (Oct 26) Read
Grok Bot added approved private certificate authorities for custom MCP servers and enterprise audit logs for trusted-certificate changes. (Oct 26) Read
Team managers can give Grok Bots shared Vault logins, while sharing controls can restrict a Team Bot to its owner. (Oct 27) Read
Reports & Data This Week
KPMG found 37% of 1,076 surveyed New Zealand consumers were open to AI assistants choosing products or services for them. (Oct 29) Read
BCG reported 59% of surveyed consumer-goods companies had no meaningful agentic AI capability, though the article did not disclose sample size. (Oct 30) Read
BCG found organisations scaling AI innovation with governance reported three times the ROI of low-innovation organisations in its global enterprise survey. (Oct 26) Read
Gartner found 78% of 297 surveyed CHROs had no alignment among leaders about where AI creates value. (Oct 26) Read
Gartner found 58% of 272 infrastructure leaders viewed scaling AI pilots into operations as an important priority over the next year. (Oct 26) Read
👋 Mike & Erin