
Kia ora! Welcome to New Zealand’s weekly roundup of AI news, events, jobs and education.
It felt like every AI heavyweight dropped a major announcement last week (check out the 🌍 Tech Updates From Global section). Google’s Gemini 3.0 model is rumoured for release on Oct 9, another big week ahead…
Hollywood is in for a shake up with the rise of Tilly Norwood (video link), the world’s first AI-generated actress, already courted by major talent agencies. Critics warn it threatens human actors while Eline Van der Velden (Creator) defends her as a creative art form rather than a replacement.
Some of you may have noticed we didn’t release an AI On The Couch episode last week. You might’ve also spotted in earlier episodes that Erin was pregnant… We hit pause to welcome our little one. Mum and baby are doing great, and we’ll be back to regular programming soon.
This week’s highlights
💼 2degrees saves 3,000 hours with AI
🌐 Composite raises $9.6M for browser AI
🏠 Home builder swaps consultants for AI
👵 Over-50s lead NZ’s AI training uplift
🎓 Kiwi unis scrap AI detectors
🎙️ Jacob Bank: the future of AI agents
💡 Tip: Find the Right AI Model For The Job
📘 The NZ AI Leadership Playbook
🏡 AI at Home: ChatGPT Scheduled Tasks
Happy reading and listening ✌️
Did someone forward you this? Sign up!
HEADLINE STORY
🎒 Kiwi kids at risk from sexual AI chatbots lacking safeguards

📣 Word On The Street: Paddy Gower investigation reveals Kiwi kids having romantic AI conversations, with no age verification required
🔎 Zooming In:
Elon Musk's xAI rolled out its Grok chatbot "Annie", who admitted to Paddy that 12-year-olds regularly use the platform.
303 deep fake reports to Netsafe last year; one in five victims are high school students.
Australia's e-safety commissioner fines platforms $50M and enforces age checks; NZ scrapped similar powers in 2024.
🏘️ Our Take: Grok’s chatbot told Paddy that 12-year-olds use the platform but are “too young to understand what I’m programmed for”. Meta ran the same playbook in August, approving bots to tell 8-year-olds “every inch of you is a masterpiece”, only reversing after Reuters exposed it. Tech giants won’t self-regulate when engagement drives profits.
Kiwi kids can access both platforms. But NZ scrapped enforcement powers in 2024 over free speech fears. Ideology won while student suicides climbed. AI platforms now operate in New Zealand with no guardrails: no compliance standards, no enforcement, no legal certainty. The incentives are backwards. Companies that ignore child safety cut costs and grow faster, while those that take responsibility carry the burden. This isn’t protecting innovation. It’s protecting negligence.
Master AI. With people who get it.
A practical Masters for working professionals. Learn in trusted cohorts, apply it at work.
academyEX is the EXceptionally proud sponsor of this week's newsletter.
💼 Business & Industry
2degrees forecasts to save 3,000 staff hours monthly by letting everyone loose on AI for 30 days. 1,500 employees attended 50+ events during August's company-wide experiment. Copilot usage spiked 400% as teams built 14 AI agents – two now in full production could deliver $2M annually in reduced churn and faster sales.
4-min read.
Our take: 2degrees showed how speed matters most in early AI adoption. They didn’t waste time designing a perfect programme; instead they opened the doors to 1,500 people across 50 events in 30 days. Most experiments failed but the ones that worked are now saving $2M a year. The real change came from letting frontline staff build their own tools. That turned resistance into ownership and created champions, and proving the lesson that AI is not about the technology. It is about creating an environment where people feel free to experiment and find efficiencies themselves.
Two 24-year-old Kiwis just raised $9.6M for browser AI that handles your busywork. Yang Fan Yun and Shine Wu's Composite automates rote tasks like updating spreadsheets and drafting emails directly in Chrome, which is 85% of office workers' daily grind. Uber, Tesla, Google and Reddit are early adopters.
4-min read.
Our take: This democratises productivity tools that were once enterprise-only. When Stanford grads and Silicon Valley VCs back two young Kiwis solving real workflow pain, it signals our tech talent can compete globally on product, not just price.
Hamilton developer replaced his consultants with ChatGPT. John Kenel slashed marketing costs 70% and pumped out 10x more content using AI across his property business. Tasks that cost $10,000 in consultant fees now run through algorithms: town planning, zoning checks, financial reports.
3-min read.
Our take: Two things: 1. This is the consultant apocalypse starting to bite. When a regional developer can cut professional service costs by two-thirds while scaling output, every billing-by-the-hour industry needs to rethink their business model and review the unit of value they deliver clients. 2. If a Property Developer can pick up and run with these tools, there’s no barrier to non-technology / IT industries running a similar playbook.
💼 Government & Legal
DIA built an AI assistant enabling conversational support that helps stressed Kiwis navigate government services without agency-hopping. Govt.nz's AI pilot used 7 language models to test which could handle users questions’ like "How do I register a death?" best. The assistant scraped selected government websites using Retrieval-Augmented Generation (RAG), then responded conversationally. Claude Sonnet and Gemini were selected, but only after prompt engineering taught them government tone and when to flag uncertainty.
7-min read.
Our take: The path forward requires two things. First, is radical honesty: show sources, flag uncertainty, let people verify everything. Second, agencies need to stop pretending they can bolt AI onto broken content infrastructure. 77-115% performance jumps came from teaching the AI to ask clarifying questions instead of guessing intent, proving the tools are powerful enough and smarter prompting beats expensive models every time.
💼 Education & Society
Over-50s now dominate AI training courses. AcademyEX reports 44% of AI course enrolments are workers aged 50+, up from mostly tech staff in their 20s a year ago. Their decades of industry experience helps them spot AI errors younger workers miss.
5-min listen.
Our take: With AI’s limited barrier to entry, experience matters more than tech fluency. A 62-year-old pharma exec can identify dodgy AI outputs that a fresh grad would trust completely. The prompt engineering might be simple, but knowing when AI's talking rubbish requires years of pattern recognition, which requires training or self-directed curiosity.
Major Kiwi unis scrapped AI detectors and accidentally fixed education. Unreliable detection tools at Massey, Auckland and Victoria caught innocent students while AI users slipped through. Universities now test what ChatGPT can't replicate: defending positions under pressure and communicating complex ideas without scripts.
3-min read.
Our take: It took AI breaking assessment systems for universities to focus on human skills. Nobody hires for essay-writing: they hire for thinking, adapting and communicating in messy situations. The degree might result in something employers want, just not what it used to certify.
🎙️ The AI Corner Podcast
This week’s guest is Jacob Bank, founder & CEO of Relay.app. Hear:
How AI agents are evolving beyond chatbots and copilots.
Why Relay pivoted through multiple prototypes to land on “agent + workflow” tooling.
Jacob’s definitions of agents vs workflows and his predictions for how junior knowledge roles will vanish by 2026.
💡 Tip of the Week: Find the Right AI Model For The Job.
Brought to you by academyEX.
With OpenRouter, you can explore and compare 500+ models from different providers, all ranked in a simple leaderboard to help you spot what model works best across areas like marketing, programming, etc.
Visit openrouter.ai
Browse the leaderboard rankings
Choose the model that fits your task
Either use the model in its usual interface (e.g. ChatGPT for GPT-5, Claude for Sonnet 4.5 etc.), or call it directly via the OpenRouter API endpoint.

📚️ AI for Business
Helping leaders and teams adapt, learn, and scale with AI.
1️⃣ The NZ AI Leadership Playbook
Tidier emails or summarised reports aren’t AI adoption, they’re the starting line. This playbook shows leaders how to move past the basics with the right tools, a 1-week test to prove value, and the mindset to embed AI into strategy.
3-min read.
2️⃣ OpenAI’s Sora 2 Goes Viral: five great clips
OpenAI’s new video model is out in limited release, and the internet already has Ronald McDonald on Love Island and Sam Altman shoplifting at Target. Beneath the memes: a serious step change in generative video.
2-min scroll with links.
3️⃣ AI is NZ’s Productivity and Scaling Advantage (If We Want It To Be)
Nine insights on why NZ’s leaders must adopt AI to stay competitive. Trust is our superpower, but shallow adoption and tall-poppy culture are holding us back.
1-min read.
4️⃣ Google vs OpenAI: The Reddit Showdown
Google quietly killed the tactic that ChatGPT was using to scrape Reddit at scale to give itself the advantage. Inadvertently, Reddit will be the major benefactor as demand for fresh, unfiltered conversations continues to grow.
1-min read.
5️⃣ Shopping Inside Of ChatGPT Is Here
OpenAI just collapsed the purchase funnel. Etsy sellers and Shopify merchants can now sell directly in chat, with Stripe powering instant checkout. Discovery → decision → transaction in one step.
1-min read.
🏡 AI at Home: ChatGPT Scheduled Tasks
ChatGPT Tasks are scheduled prompts that run automatically and notify you in the app or on desktop.
Brilliant interactive checkpoints that you can work with inside ChatGPT rather than using a to-do list app or phone alarms for reminders.
The AI does part of the work: asking questions, generating summaries, or creating content right when you need it.
Each task is maintained in the same conversation thread, so you’ve got the full history to look back on (makes keeping journalling habits easy).
I use the microphone on my smartphone keyboard to effortlessly speak instead of typing, which enables me to brain dump on the go.
Sample prompt framework: “Schedule for me a task every [WHEN] to [ACTIVITY]”.
They’re not yet advanced, but with the launch of Pulse, growing list of connectors, and new capabilities improving, these will only get better. For now, they’re a great way to start small when you want to lean on ChatGPT for a regular task.
A couple of examples that I regularly use:

🌍 Tech Updates From Global
The selected top headlines from each major AI tech company.
OpenAI
Launched Sora 2, a video generation app with social media features, now invite-only in the UK, US and Canada.
Added ChatGPT features including Pulse for Pro users, “Buy Now” buttons and Instant Checkout powered by Stripe, enabling in-chat transactions for Etsy and Shopify (soon).
Became the world’s most valuable private company at a $500B valuation following a $6.6B secondary sale.
Released upgraded Gemini models enabling robots to execute multi-step tasks and retrieve real-time online information.
Launched AI radio hosts through Google Labs for trivia and contextual commentary between music tracks.
Expanded ‘Help Me Edit’ in Google Photos to all US Android phones, supporting voice and text edits powered by Gemini AI.
Deployed Gemini for Home as the backbone of its smart home ecosystem, integrated into new devices and services.
Introduced Dreamer 4 by DeepMind, enabling AIs to train via mental simulations.
Anthropic
Released Claude Sonnet 4.5, capable of coding uninterrupted for 30+ hours, including Code Checkpoints and Agent SDK tools.
Announced global expansion with 100+ roles in Dublin, London and Zurich.
Promoted context engineering as a superior method to prompt engineering for agent design.
Reported that 80% of Claude users are now outside the United States.
Microsoft
Rolled out Agent Mode in Excel and Word, allowing natural language-driven productivity and “vibe working” collaboration.
Launched Microsoft 365 Premium with Copilot integration for £19.99 per month, competing directly with ChatGPT Plus.
Meta
Announced plans to integrate AI chat data into ad targeting, starting 16 December.
Acquired Rivos, an AI chip start-up, to reduce dependence on NVIDIA and boost internal chip R&D.
NVIDIA
Powered the AI-enabled 2025 Ryder Cup tournament, improving fan experience and logistics.
Apple
Announced ‘Manzano’, a unified system for image understanding, generation and Siri enhancement.
Perplexity
Launched AI-native Comet browser to the public for free (previously only available to paying subscribers or invitees).
California (Regulatory)
Passed a new AI safety transparency law, mandating disclosure of safety protocols and incident reporting by AI companies.
📅 AI Events in New Zealand
13 AI events across the country this week.
This week’s featured event:
Snowflake World Tour, Wed - Auckland: Snowflake brings AI and data innovation to Auckland for data leaders and tech professionals.
📅 Promote your event with us. Reply to let us know.
💼 AI Roles Around Aotearoa
Picklist of 🌶️ HOT 🌶️ new roles in AI this week.
AI Solution Specialist - Manager, PWC: Wellington
💼 Promote your job with us. Reply to let us know.
🤦 ️ AI Fail Of The Week
We all love AI, but it’s certainly far from perfect 🤔 …

Even AI gets lazy….
👋 Mike & Erin