In partnership with
Kia ora! Welcome to New Zealand’s weekly roundup of AI news and education.
“AI doesn’t understand our business”. “I keep correcting the same mistakes”. “Everyone in the team gets different results”.
I hear these frustrations constantly. Working with businesses keeps reinforcing my view that people need more practical education in setting AI up around their work, including the context it needs and how corrections become reusable instructions.
For example, I made this video using ChatGPT in about 10 minutes on Friday. That speed reflects the operating system I’ve built around it: stored examples, preferences and rules that my workflow draws on each time. The voiceover and synchronisation need work, and my feedback will help improve the next attempt.
Screenshot of animated video created using ChatGPT.
My article on raising a team’s AI watermark explains how to extend that approach across a business: move from isolated chats to shared context, then make approved improvements available to everyone. When one person fixes a recurring mistake, their colleague should benefit next time.
Want help building your setup? My four-week 1:1 AI Coaching Programme works around your tools and actual workflows. Reply for the details.
Happy reading ✌️
Did someone forward this? Sign up.
Webinar results: we’ve heard you
We’ve heard you. Thanks to the 157 readers who voted in last week’s webinar poll.
Turning a spreadsheet into a report with checked numbers was the top choice, with 67 votes (42.7%). We’ll get a practical webinar organised on that topic and share the details here once confirmed.
| Spreadsheet → report with checked numbers | ||
| 67 · 42.7% | ||
| Scattered documents → decision brief | ||
| 43 · 27.4% | ||
| Meeting transcript → proposal or presentation | ||
| 25 · 15.9% | ||
| Long email thread → action plan | ||
| 17 · 10.8% | ||
| Wouldn’t attend a webinar | ||
| 5 · 3.2% | ||
Which practical AI webinar would you attend? 157 votes. Results as at 4 October 2026.
🇳🇿 New Zealand News
 | Seymour says his office has tested AI in policy development. He supports AI-drafted Cabinet papers with ministerial review, while arguing New Zealand should follow overseas regulatory decisions. |
Our take: What were those three years actually buying? Seymour’s AI example doesn’t prove it solved the policy problem, but it should make government explain where the time goes. If it buys evidence, consultation and better decisions, protect it; if it buys repeated drafting and internal queues, the delay gets harder to justify. There’s a nice twist here too: the technology outpacing the regulator might also help it catch up.
 | NZ writing startup Capiche is seeking commercial partners and investors. The council profile says its AI assists expression rather than generating students’ ideas, and reports 1,000 users and engagement with 75 schools. |
Our take: How much intelligence does your business overlook because it doesn’t arrive in a polished paragraph? The person with the slickest presentation isn’t always the person who understands the problem best. Capiche’s decision to help people express their own ideas could give more of that knowledge a hearing. It still needs to prove it works, but helping more people contribute is an opportunity worth backing.
 | A teachers’ union is warning against replacing human judgement in AI marking. The PPTA conference discussion seeks teacher responsibility for assessment, rather than reporting a new NZQA automated grading rollout. |
Our take: The framing here feels too binary. Some routine marking tells a teacher very little, and automating it could free time for better lessons and the kids who need attention. More complex assessment needs the teacher involved, because doing less marking can also mean knowing less about a student’s progress. The question is where that trade-off leaves students better off.
 | New Heart Foundation funding includes AI analysis of heart scans. The Auckland research sits within more than $6 million awarded across 38 research and training recipients, so that total is not an AI budget or a proven treatment result. |
Our take: It’s refreshing to see AI news about work that could improve people’s lives. These stories deserve more attention alongside the warnings and controversies. The next thing to watch is whether better scan analysis helps clinicians get people into the right treatment sooner, because that’s where the benefit reaches patients.
 | AI agents could help New Zealand firms find suppliers and respond to shipping disruption. University of Auckland professor Ilan Oshri argues procurement teams need better data and internal expertise alongside the software, drawing on an international study of more than 200 professionals. |
Our take: The useful test is what happens when a supplier fails. Can the team find and check alternatives faster, with agreed spending limits and someone responsible for the final choice?
 | Labour’s Reuben Davidson wants a more active government role in AI. In separate RNZ and interest.co.nz interviews, he outlines an AI office and minister, workforce planning, copyright protections and renewable-energy requirements for data centres. A firm spending estimate remains unspecified. |
Our take: An AI office needs a clear job beyond writing policy. Businesses need to know which decisions it would make, what support it would provide and how quickly it could respond.
 | Sterling gives its AI finance autopilot a familiar face. The NZ startup’s new brand uses staplers, paperclips and calculators to introduce software that handles invoices, bank reconciliation and month-end routines, while logging its actions. |
Our take: It looks like stationery, but you’re effectively hiring it into your finance team. The familiar branding is clever; the bigger question is what you’ll let it do with your money and records. Give it clear limits, someone to check its work and a way to ask for help. A log of its actions won’t tell you whether they were sensible.
 | NZ experts examine the prospect of AI data centres in space. Science Media Centre’s expert responses to Google’s orbital computing experiment raise questions about cooling, launch weight, communications and debris. This is a proposal under investigation, rather than a NZ deployment. |
Our take: Imagine it becoming easier to launch a data centre into space than get it connected to the grid. We’re not there yet, but how much of the hold-up on Earth could we fix? It echoes Seymour’s point about government keeping up. Moving into orbit won’t make the responsibilities disappear either, because debris and interference become everyone’s problem.
Your AI prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning (constraints, edge cases, examples, tone) and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
Used by teams at Microsoft, Amazon, and Notion. Try Wispr Flow free on Mac, Windows, iPhone, and Android.
📚️ Mike's Takes
Helping leaders and teams adapt, learn, and scale with AI.
1️⃣ One AI interface can connect the work after a meeting: A transcript can feed a debrief, coaching, content ideas, CRM updates and a branded proposal through one interface. Connectors and computer use join the steps together, while business context, permissions and review determine whether the workflow is useful.
Watch.
2️⃣ An always-on agent needs the context built before it arrives: Mike explores how ChatGPT Dots could monitor connected messages, revise plans and investigate recurring problems. Teams with documented priorities, business rules and approval boundaries already have useful context to give an agent. Access and real-world performance still need testing.
Read.
3️⃣ OpenAI uses Sachin Tendulkar to sell ChatGPT like a mass consumer brand: The partnership adds a familiar face to OpenAI’s wider push into India, alongside education, enterprise and infrastructure deals. Mike looks at how cricket and celebrity endorsement could make personal AI assistants familiar to a much broader audience.
Watch.
4️⃣ Keep business context portable as the AI assistant race accelerates: Microsoft, Meta and other providers are approaching delegated work through different interfaces. Once an assistant knows a business’s routines, switching gets harder. Keeping Skills, rules and context under the business’s control makes it easier to move when better options emerge.
Read.
🌍 Global Tech Updates
The selected top headlines from each major AI tech company.
Anthropic
Sonnet 5.5 launched with million-token context, with Anthropic reporting 30% faster operation and up to 30% lower costs than Sonnet 5. (Sep 28) Read
Claude Code Mods let teams customise prompts, tool behaviour and interface elements using TypeScript plugins in the CLI and desktop app. (Oct 1) Read
Claude Code added command-line plugin configuration and installation-time connector settings, reducing manual setup steps for repeatable team installations. (Sep 29) Read
Claude Code reviews gained a configurable findings limit, letting teams request more issues or narrow the review without changing the workflow. (Oct 2) Read
Claude Code now displays actual dollar spending against gateway limits in usage views, giving administrators clearer visibility into monthly consumption. (Sep 28) Read
Claude for Government reached general availability for US federal and state agencies, delivering coding and agent capabilities through FedRAMP High infrastructure. (Sep 30) Read
Anthropic committed $100 million to train 10,000 engineers through Claude Frontier Academy, with Commonwealth Bank among the first participating organisations. (Oct 2) Read
Anthropic found GLM-5.3 safeguards could be bypassed in simulated cyber misuse tests, highlighting risks as advanced capabilities reach downloadable models. (Sep 29) Read
OpenAI
GPT-6.1 Sol launches for coding and professional work, with standard input and output token prices one-fifth of Astra’s. (Sep 29) Read
Astra Ultrafast arrives in the API and Pro 500, offering a faster paid service tier for time-sensitive work. (Sep 29) Read
The Agents API adds hosted-browser computer use, allowing applications to delegate website tasks with access approvals and sign-in controls. (Sep 29) Read
Dots launch as always-on agents with their own cloud computers, taking ongoing responsibilities for eligible Pro and Business Premium users. (Sep 29) Read
Codex Cloud introduces reusable development environments, letting coding tasks continue in isolated workspaces while the user’s computer sleeps. (Sep 29) Read
ChatGPT Space brings files, Pages and shared project material into a collaborative workspace for eligible Pro, Business and Enterprise users. (Sep 29) Read
Pages introduce collaborative documents where people and agents can edit text, add charts and build interactive content together. (Sep 29) Read
The Meetings plugin enters macOS beta, saving personalised meeting summaries and action items to Space for Pro and Business users. (Sep 29) Read
ChatGPT adds virtual clothing try-ons and saved shopping favourites, with reference-photo controls available on mobile and the web. (Oct 1) Read
ChatGPT’s iOS camera adds multi-page scanning, automatically combining photographed notes or documents into one PDF for a conversation. (Oct 1) Read
Finances expands to US Free and Go users, connecting financial accounts for personalised spending, saving and investment explanations. (Oct 2) Read
Codex Security Cloud adds scheduled repository scans and proposed fixes, with access to Daybreak Blue models through the service. (Sep 29) Read
Plugin extensions bring interactive panels, sidebar experiences and custom file viewers into ChatGPT for developers building native workflows. (Sep 29) Read
Business and Enterprise teams can share tasks and delegate recurring work, jointly refining instructions for scheduled or event-triggered actions. (Sep 29) Read
ChatGPT enters Slack and Microsoft Teams conversations, using approved connected tools when colleagues mention it in channels or messages. (Sep 29) Read
Eligible Plus and Pro users can authorise participating third-party applications to spend their ChatGPT allowance, with per-app usage controls. (Sep 29) Read
Private Intelligence introduces zero-data-retention safety processing, with a confidential-computing Private Inference preview planned for later this year. (Sep 29) Read
Gemini 4 Argon begins restricted rollout to trusted cyber defenders, with broader developer, enterprise and consumer access planned after testing. (Sep 30) Read
Gemini skills introduce reusable instructions that activate from matching prompts, with multiple skills combinable in the same conversation. (Sep 30) Read
Workspace skills begin rolling out on 5 October, with a staged Gems migration and eventual retirement timetable announced. (Sep 30) Read
Google Vids adds custom caption fonts, positioning and animation, helping teams match generated videos to their own brand style. (Sep 30) Read
Docs, Sheets and Slides APIs add programmatic comments, while Docs gains suggested edits for automated review workflows. (Sep 30) Read
Chinese Frontier Labs
QwenCloud added Qwen-Image-2.1-Pro, combining image generation, editing, transparent output and up to ten reference images in one service. (Oct 2) Read
DeepSeek released DeepGEMM Ascend, bringing compatible inference kernels to Huawei Ascend 950 hardware while preserving existing APIs and development workflows. (Sep 30) Read
Open-Weight Model Updates
Cloudflare open-sourced Clef and Clef-flash decision models under Apache 2.0, adding visual classification and reinforcement-learning customisation through Workers AI. (Oct 1) Read
Microsoft and GitHub
GitHub Copilot added computer-use preview on Windows and macOS, letting agents operate desktop applications with user approval. (Oct 1) Read
Dynamic workflows reached Copilot CLI, app and SDK, combining programmed steps, parallel agents and review checkpoints into repeatable processes. (Oct 1) Read
Microsoft Copilot added GPT-6.1 Sol and Claude Sonnet 5.5, extending model choice across its workplace AI experiences. (Sep 30) Read
Copilot’s plugin registry lets administrators approve plugins centrally and make their skills and connectors available across Microsoft workplace applications. (Sep 30) Read
New Copilot Business subscriptions through partners will default to usage billing from December 1, with adjustable monthly credit limits. (Oct 1) Read
Meta
xAI
Team Bots launched as shared AI colleagues, combining reusable team skills with private conversations and separate memories for each user. (Sep 28) Read
Team Bots gained individual Slack identities, allowing colleagues to invite an AI teammate into channels and collaborate around shared work. (Sep 28) Read
Grok Voice Transcribe 1.0 reached end of life, with requests automatically routed to version 2.0 at unchanged pricing. (Oct 2) Read
👋 Mike & Erin